Azure Content Understanding agentic mode
Agentic mode in Azure Content Understanding is now in preview for multi-step document analysis.
The Deploy Log | Free, no signup
Models and APIs
393 rows in Models and APIs in The Deploy Log, newest first. They are listed from to , each by its publisher's date or, where the record carries none, its edition's date. Every item sourced to the publisher's own page.
Claude Fable 5.2 on OpenRouter
Anthropic Claude Fable 5.2 is available on OpenRouter at $8/$40 per 1M tokens with 1.0M context.
Google Maps Platform in AI Studio
Google Maps Platform is now available directly inside Google AI Studio at no cost.
Meta Muse Ember
Meta Muse Ember is available on OpenRouter at $0.10/$0.50 per 1M tokens with a 1M-token context window.
Qwen3.8-27B-DSpine-b16
Tencent released a five-layer draft model for Qwen3.8-27B with block size 16 and an 8,192-token sliding window.
Arete-27B
Nous Research released Arete-27B, a 27B model post-trained over Qwen3.8-27B, on Hugging Face.
Cola2
Cola2 continuous-latent language model checkpoint released on Hugging Face, trained on 346.03B tokens.
ElevenLabs Python SDK v2.77.0
Adds agent workflows as code and an auth parameter to speech engine create and update.
LiquidAI LFM3-1.7B
LiquidAI LFM3-1.7B is available on OpenRouter at $0.08 input and $0.20 output per 1M tokens.
Sage-SciPerception
Sage-SciPerception released on Hugging Face, connecting scientific encoders to language models via trainable projectors.
Show the other 383 rows
SageMaker Multi-Instance GPU
SageMaker Inference supports Multi-Instance GPU to maximize GPU utilization for model endpoints.
googleapis/python-genai v3.0.0
Python GenAI SDK v3.0.0 defaults Gemini Developer API to v1 instead of v1beta.
java-genai v2.0.0
java-genai v2.0.0 changes the default Gemini Developer API version from v1beta to v1; Vertex AI defaults unchanged.
Cross-Layer-TopK-CLT
ByteDance Seed released a 94.2B parameter cross-layer transcoder for Qwen3-8B-Base on Hugging Face, 377 GB total.
Gemini Omni 1.2 Flash
Google's Gemini Omni 1.2 Flash video generation model is available on OpenRouter at $1.50/$9 per 1M tokens.
Gemini Omni 1.2 Flash
Gemini Omni 1.2 Flash supports 15-second video clips and lower pricing.
Ling 3.0 Tiny
InclusionAI Ling 3.0 Tiny is on OpenRouter at $0.015/$0.045 per 1M tokens with 262K context.
ManiFlow-TB
Allen Institute for AI released ManiFlow-TB, a 322.5M parameter robot action model, on Hugging Face.
Microsoft Foundry October updates
Browser Automation is generally available and the Decisions API is in Foundry; the Microsoft Copilot harness is in public preview.
Molmo2-4B-tb
Allen Institute for AI released Molmo2-4B-tb, a 4.5B parameter vision-language model fine-tuned for robot block-stacking, on Hugging Face.
Qwen3-30B-A3B-MoE-SAE
Sparse autoencoders for Qwen3-30B-A3B MoE outputs at layers 4, 30, and 46, with 2^20 latents each.
ElevenLabs Python SDK v2.76.0
Adds entity_redaction to the realtime speech-to-text wrapper in the ElevenLabs Python SDK.
Embed 5 and Rerank 5 on OpenRouter
Cohere's Embed 5 and Rerank 5 models are now available on OpenRouter, including Pro and Fast variants.
Gemini Omni 1.1 Flash
Google's Gemini Omni 1.1 Flash video generation model is available on OpenRouter at $1.50/$9 per 1M tokens.
Ripple 65B Preview
Ripple-65B-Preview, a 65B model with native tool calling, is available on OpenRouter at $0.20/$0.60 per 1M tokens.
Tencent Hy4 on OpenRouter
Tencent Hy4 is available on OpenRouter at $0.417/$1.251 per 1M tokens with 1M context.
cohere-python 7.3.1
cohere-python 7.3.1 regenerates the SDK via Fern CLI.
cohere-python 7.3.2
cohere-python 7.3.2 fixes argument forwarding through patched get_headers and regenerates the SDK.
ElevenLabs SFX v3
SFX v3 (eleven_text_to_sound_v3) is now available to every workspace, with better prompt following and natural sound.
GLM 5.3 FlashX Legal
Z.ai GLM 5.3 FlashX Legal is available on OpenRouter at $0.37/$1.25 per 1M tokens with 1M context.
Kimi-K3.1-Pro
Moonshot AI released Kimi-K3.1-Pro, a 2.8T-parameter open-weight multimodal model with 1M context, on Hugging Face.
ChatGPT on Bedrock GovCloud
The ChatGPT desktop app is now available on Amazon Bedrock in AWS GovCloud (US), routing Codex model inference through Amazon Bedrock Mantle in us-gov-west-1 with IAM governance and zero data retention support.
Mistral Large 4 (Le Chonk)
Mistral Large 4, a 1T-parameter MoE model with 49B active parameters and 1M context, is on Hugging Face.
java-genai v1.80.0
Adds continuation_token, video upscale, and video understanding.
js-genai v2.32.0
Adds continuation_token, video upscale, and video understanding.
python-genai v2.34.0
Adds continuation_token, video upscale, video understanding, and Python 3.15 support.
LTX-2-Marking
PixelSeal and AudioSeal marking weights and configs for LTX-2 are available on Hugging Face.
NVIDIA Nemotron 3.5 Transcribe
Open-weight model for multilingual transcription, speaker attribution, translation, and audio Q&A, now available on Hugging Face and ModelScope.
Particle-1 on OpenRouter
Particle-1 is available on OpenRouter at $0.25 per 1M input tokens and $1 per 1M output tokens with 1.0M context.
GLM-5.3-Flash-singprobe
Released an 8.13M parameter streaming guardrail probe for GLM-5.3-Flash with under 0.5% decode overhead.
Ling-3.1-flash-1019
Ling-3.1-flash-1019 released with 560B total parameters, 25B active, 1M context.
Metatron-v1 ONNX exports
ONNX inference exports of the unified Metatron G1 scene/recovery policy, with demo and racer files.
Qwen3.8-Flash-Next-singprobe
Released a 5.18M parameter streaming guardrail probe for Qwen3.8-Flash-Next with under 0.5% decode overhead.
Azure AI Speech LLM 09-2026
Adds 25 languages including 11 Indic locales and claims up to 2x transcription speed.
GPT-6.1 Terra
GPT-6.1 Terra is available with standard API prices 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
GPT-6.1 Terra
GPT-6.1 Terra is available on OpenRouter at $1/$5 per 1M tokens with 1.1M context.
GPT-6.1 Terra Pro on OpenRouter
OpenAI GPT-6.1 Terra Pro is available on OpenRouter at $1/$5 per 1M tokens with 1.1M context.
LFM3-1.7B-Base
Liquid AI released LFM3-1.7B-Base, a 1.73B parameter pre-trained text model with 256K context, on Hugging Face.
MiniCPM-OCR-1.0-DSpark
OpenBMB released MiniCPM-OCR-1.0-DSpark, a 296M parameter draft head that speeds up MiniCPM-OCR 1.0 decoding by 55%.
Unlimited-OCR-2
Unlimited-OCR-2 released on Hugging Face with SR-SWA for unlimited-horizon document parsing.
Ling-3.1-flash-base
Ling-3.1-flash-base released on Hugging Face and ModelScope with 560B total and 25B active parameters.
MS-Human-700
MS-Human-700 whole-body musculoskeletal model with 700 muscle-tendon units, MuJoCo models, Gymnasium environments, and DynSyn-SAC training code.
Hy-Nomas
Tencent released Hy-Nomas BF16 Diffusers weights on Hugging Face, a video world generation model with 4-step distilled inference.
Atria-Dawn
InternLM released Atria-Dawn, a 744B-parameter MoE agentic model with 512K context, on Hugging Face and ModelScope.
Kimi K3.1 Pro
MoonshotAI Kimi K3.1 Pro is available on OpenRouter at $2/$10 per 1M tokens with 1M context.
MotionBricks
NVIDIA released MotionBricks pretrained SafeTensors weights for SOMA and Unitree G1 on Hugging Face.
Nova-3 Multilingual
Nova-3 multilingual now transcribes 55 languages with one model and adds language_hint.
java-genai v1.79.0
java-genai v1.79.0 adds store to GenerateContentConfig, labels to LiveConnectConfig, and Live API in-session context updates.
js-genai v2.31.0
Adds store to GenerateContentConfig, Live API in-session context updates, and generation_output_config.
Amazon Polly Generative engine
The Amazon Polly Generative engine upgrade adds 14 new voices across 9 locales, including general availability of Hindi and Finnish for the first time, and is a drop-in upgrade with unchanged voice names and API calls.
Decisions API
Decisions API is generally available, built with GPT-6 Luna, up to 10x faster than Responses API, with 400 questions per request.
ThinkingBox-Training
ThinkingBox-Training is now available on GitHub, with Qwen3.8-27B average success rising from 51.70% to 60.78%.
WeVisOne-VL-4B and WeVisOne-VL-2B
Tencent released WeVisOne-VL-4B and WeVisOne-VL-2B on Hugging Face with a 9.6M-parameter pixel decoder.
cohere-python 7.3.0
cohere-python 7.3.0 adds a relevant_spans parameter for Rerank models.
python-genai v2.32.0
python-genai v2.32.0 adds base_voice, generation_output_config, and Live API in-session context updates.
AWS Agent Registry
AWS Agent Registry lets agents discover tools at runtime.
OpenAI GPT models on Bedrock in Canada
OpenAI GPT models on Amazon Bedrock are available from Canada with cross-Region inference.
PyTorch 2.15 MPS
PyTorch 2.15 adds FP8 on MPS, 5-6.5x faster MoE decoding, and faster LLM decoding paths.
Voxtral 3
Voxtral 3 voice agent SDK and speech models in 50+ languages, available via EU-hosted API or self-deployment.
Browser automation tool in Microsoft Foundry
Browser automation tool in Microsoft Foundry is now generally available, giving agents a managed browser for website workflows.
Cosmos3-H-Dreams
NVIDIA released Cosmos3-H-Dreams, a 15.17B real-time surgical world model, on Hugging Face.
Cosmos3-H-Embodied-Nano
NVIDIA released Cosmos3-H-Embodied-Nano, a 16B surgical robotics model, on Hugging Face.
Pegasus 1.5 in Amazon Bedrock
TwelveLabs Pegasus 1.5 is now available in Amazon Bedrock.
Structured outputs for Claude on Bedrock
Structured outputs for Anthropic Claude models on Amazon Bedrock are now available.
cosmos3-nano-libero10-sdg-export-D2_s43
NVIDIA released a Cosmos3-Nano LIBERO-10 policy checkpoint, about 30 GB, loadable by the cosmos-framework policy server.
cosmos3-nano-libero10-sdg-export-S5_2_s43
NVIDIA released a Cosmos3-Nano LIBERO-10 policy checkpoint, about 30 GB, loadable by the cosmos-framework policy server.
Claude structured outputs on Bedrock
Structured outputs for Anthropic Claude models are available on both bedrock-runtime and bedrock-mantle endpoints.
Android Earthquake Alerts
Android Earthquake Alerts has expanded from 98 countries and regions to more than 200, after sending over one billion alerts since 2021.
Nemotron 3.5 Super VL
NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter model that is 5x faster than the next best in its class, alongside Nemotron 3.5 Bolt 74B-A9B 24GB for local agentic workloads.
Timestral
Mistral released Timestral, a family of time-series foundation models from 5M to 209M parameters that ranks first on GIFT-Eval, TIME and fev-bench and runs on CPUs without a GPU.
Muse Spark Open
Meta's Muse Spark Open, an 816B-parameter mixture-of-experts coding model with a 1 million-token context window, is now available in Amazon SageMaker JumpStart in BF16 and NVFP4 variants.
DeepSeek V4.1 Flash 1017
DeepSeek V4.1 Flash 1017 on OpenRouter at $0.15/$0.60 per 1M tokens, 1M context.
DeepSeek-V4.1-Flash-1017
DeepSeek-V4.1-Flash-1017 released on Hugging Face with MIT license, 1M context, and max 256K tokens.
Elorian Glimpse
Elorian Glimpse multimodal visual reasoning model is available on OpenRouter at $0.60/$3 per 1M tokens with 262K context.
Beam
Beam, a 501B-parameter open MoE model with 1M context, is out on Hugging Face under Apache 2.0.
Muse-Spark-Open-816B-A42B-DFlash2
NVIDIA released a 4.38B-parameter DFlash 2 draft checkpoint for Muse-Spark-Open-816B-A42B on Hugging Face.
NVIDIA Nemotron Labs Dory1.0 4B
NVIDIA released the 4B-parameter Dory1.0 model in BF16 with up to 256K context on Hugging Face.
NVIDIA-Nemotron-Labs-Dory1.0-4B-MXFP8
NVIDIA released Dory1.0-4B-MXFP8, a 4B parameter looped LLM with 256K context, on Hugging Face.
Qwen3-8B-DSpine-b16
Tencent released Qwen3-8B-DSpine-b16 on Hugging Face with inference code.
Reflection Beam 1.0 501B A23B
Reflection Beam 1.0 501B A23B is available on OpenRouter at $0.25 input and $1 output per 1M tokens with 1M context.
Apertus 1.5
Apertus 1.5 8B and 70B weights released with image/audio input, 262,144-token context, and thinking mode.
DeepSeek-V4.1-Flash-bf16
Prime Intellect released a bfloat16 re-export of DeepSeek-V4.1-Flash on Hugging Face, 1.34 TB across 48 safetensors shards.
ElevenLabs Python SDK v2.73.0
ElevenLabs Python SDK v2.73.0 adds create_and_wait helpers for Flows and type fixes.
Hy-Memory-30B-A3B
Tencent released Hy-Memory-30B-A3B, a 3B active parameter memory writer model, on Hugging Face.
Meta Muse Spark Open
Meta Muse Spark Open is available on OpenRouter at $1.25 per 1M input tokens and $4.25 per 1M output tokens.
Meta Muse Spark Open on Together AI
Meta Muse Spark Open is now live on Together AI.
Moondream 3 Preview
Moondream 3 Preview, a 9B-parameter vision-language model with 2B active parameters, is available on OpenRouter at $0.30/$2.50 per 1M tokens.
Moondream 3.1 on OpenRouter
Moondream 3.1 is available on OpenRouter at $0.30 per 1M input tokens and $2.50 per 1M output tokens.
Muse Spark Open
Meta's open-weight multimodal mixture-of-experts model with ~816B total parameters, ~42B active per token, and up to 1M context is available on Hugging Face.
NVIDIA Nemotron 3.5 Super VL
NVIDIA Nemotron 3.5 Super VL is available on OpenRouter at $0.15/$0.50 per 1M tokens with 1M context.
Nemotron-3.5-Rerank-8B-BF16
NVIDIA released Nemotron-3.5-Rerank-8B-BF16 on Hugging Face, an 8.38B parameter multilingual and multimodal reranker for retrieval pipelines.
scFM-dataselection checkpoints
572 pretrained single-cell model checkpoints and PCA components, approximately 32.1 GB, on Hugging Face.
FLUX 3 Image Klein Prompt Upsampler
FLUX 3 Image Klein Prompt Upsampler, a 9B vision-language model, released under Apache 2.0 for expanding image prompts.
FLUX 3 Image Klein for ComfyUI
FLUX 3 Image Klein, a 5.6B open-weights text-to-image and editing model, released with ComfyUI files in BF16, int8, and NVFP4.
Grok Voice 3.0
Grok Voice 3.0 is available with speech-to-speech at the same price as Grok Voice 2.0, and grok-voice-latest now points to it.
UserRepGen-4B
Microsoft released UserRepGen-4B, a fine-tuned Qwen3.5-4B model for generating user representations, on Hugging Face under MIT license.
java-genai v1.78.0
java-genai v1.78.0 adds gemini-4-argon to the Model enum and fixes grounding metadata merging for automatic continuation.
ElevenLabs Python SDK v2.72.0
ElevenLabs Python SDK v2.72.0 adds iterable audio chunks support and close code in RealtimeEvents.CLOSE.
Gemini 4 Argon
Gemini 4 Argon is now available in the Gemini app.
Grok 4.7 on Amazon Bedrock
Grok 4.7 is available on Amazon Bedrock.
Ling-3.1-flash
Ling-3.1-flash released with 560B total parameters, 25B active, 1M-token context, on Hugging Face, ModelScope, and OpenRouter.
LiquidAI LFM2.5-24B-A2B
LiquidAI LFM2.5-24B-A2B is available on OpenRouter at $0.10/$1 per 1M tokens with 128K context.
Pareto 26.9
Pareto 26.9 is available on OpenRouter at $2.50/$7.50 per 1M tokens with 262K context.
ScaleDown Extract
ScaleDown Extract is available on OpenRouter at $0.04 per 1M tokens with 1.0M context.
ScaleDown Summarize
ScaleDown Summarize is available on OpenRouter at $0.04 per 1M tokens with 1.0M context.
java-genai v1.77.0
Adds automatic continuation to generateContent and chat sessions, billable token counts, and transcription offsets.
Meta Muse Wonder on OpenRouter
Meta Muse Wonder is available on OpenRouter at $4/$20 per 1M tokens with a 1M context window.
OpenAI GPT models on Amazon Bedrock
OpenAI GPT models on Amazon Bedrock with global cross-Region inference in Thailand, Malaysia, Singapore, Indonesia, and Taiwan.
Trident-Qwen3VL-2B
OpenBMB released Trident-Qwen3VL-2B, a 2B multimodal embedding model for mixed-modality retrieval, on Hugging Face.
SynthID Detector
Google expanded SynthID Detector globally in English, letting anyone check if an image, video or audio was made with AI from Google or partners including OpenAI, NVIDIA, Kakao and soon Apple, after watermarking over 180 billion images and videos and 240,000 years of audio.
MAI Code 1.1 Flash
Microsoft AI developed a local version of MAI Code 1.1 Flash, a 137 billion total and 6.8 billion active parameter mixture-of-experts model, with the quantized on-device version coming in at 53GB, an 80% reduction in size, and scoring 70.80% on SWE-Bench Verified.
Sophos Daybreak
Sophos cut average response time for cases using agents from approximately 38 minutes to 89 seconds, a 96% reduction, and now resolves 52% of MDR cases end-to-end with AI built through OpenAI Daybreak.
Asana browser agent
Asana used GPT-6 Astra in Codex to optimize StackAI browser agent on GPT-6.1 Sol, cutting estimated model cost 76x to $0.47 per run and making runs 5x faster at roughly four minutes, with 89% of input from cache at 5% of the uncached price.
Agile One S SSD Place
NVIDIA released a GR00T N1.7 SSD placement deployment model with three cameras on Hugging Face.
Agile One S Walk GR00T N2
NVIDIA released the agile_one_s_walk_n2_1680 walking deployment model checkpoint on Hugging Face.
Qwen-Image-2.1-Turbo
Qwen-Image-2.1-Turbo, an accelerated 7B text-to-image and image editing model with 8 denoising steps, is available on Hugging Face.
Anthropic Cyber Mission
Anthropic launched the Cyber Mission with the Critical Infrastructure Defense Program and free OSS Scanner for open-source projects.
LIDstral-Arabic
Mistral released a 51-class Arabic language and dialect classifier on Hugging Face with download and Python usage instructions.
Midjourney Thinking Mode & Shared Folders
Thinking mode reruns 8.2 jobs to fix missed prompt details, and folders can now be shared with collaborators or viewers.
StepFun Step 5 Preview
StepFun Step 5 Preview is available on OpenRouter at $1/$2.70 per 1M tokens with 1M context.
Voxtral-Mini-4B-Realtime-Arabic
Mistral released a 4.4B parameter streaming Arabic speech-to-text model with 8.82% average CER at 480 ms delay.
Claude Haiku 5.5 on OpenRouter
Claude Haiku 5.5 is live on OpenRouter at $0.10/$0.50 per 1M tokens with a 1M context window.
Java GenAI SDK v1.76.0
Adds lyria-3.5, gemini-omni-1.1-flash, and gemini-omni-flash-preview to Interactions Model enum, plus continuation_token support.
Nova-3 Medical Improved Models
Improved Nova-3 Medical English models for batch and streaming with fewer word errors and fewer errors on medical entities.
Google Flow Music Spaces
Spaces in Google Flow Music lets creators build custom instruments or effects.
Google Nano Banana 2.1
Google Nano Banana 2.1 image generation and editing model is available on OpenRouter at $1.50/$30 per 1M tokens.
Mistral Large 4 on OpenRouter
Mistral Large 4 is available on OpenRouter with a 512K context window and 256K output tokens.
Nova-3 Medical Multilingual
Nova-3 Medical now supports multilingual transcription across 10 languages for batch and streaming.
GLM 5.3 on Amazon Bedrock
GLM 5.3 is now available on Amazon Bedrock.
Nova-3 Improved Models
Improved Nova-3 monolingual models for Catalan, French, Indonesian, and Vietnamese are live for all users.
Qwen3-0.6B-Reverse-Text-SFT-untied
Untied copy of Qwen3-0.6B-Reverse-Text-SFT with tie_word_embeddings false for trainers that don't support tied LM heads.
Qwen3-0.6B-untied
Untied copy of Qwen3-0.6B with tie_word_embeddings false for trainers that don't support tied LM heads.
Qwen3-4B-Instruct-2507-untied
Untied copy of Qwen3-4B-Instruct-2507 with tie_word_embeddings false, loadable by trainers that don't support tied LM heads.
Qwen3.5-4B-untied
Untied copy of Qwen3.5-4B with tie_word_embeddings false, loadable by trainers that don't support tied LM heads.
pplx-decider-v1.1-27b
Perplexity released pplx-decider-v1.1-27b on Hugging Face, raising Decision Index from 56.4 to 61.56.
Qwen3.5-0.8B-Reverse-Text-RL
Prime Intellect released Qwen3.5-0.8B-Reverse-Text-RL, a short RL fine-tune for the reverse-text environment.
Qwen3.5-0.8B-Reverse-Text-SFT
Prime Intellect released Qwen3.5-0.8B-Reverse-Text-SFT, a short SFT fine-tune of Qwen3.5-0.8B.
MAI-Transcribe-2-Streaming
Microsoft launched MAI-Transcribe-2-Streaming, which debuts at no. 1 on Artificial Analysis, alongside MAI-Voice-2.1 and MAI-Voice-2.1-Flash for speech generation.
LFM2.5-350M-Diffusion-Exp
Liquid AI released LFM2.5-350M-Diffusion-Exp, a 350M block-diffusion model on Hugging Face with inference code and SGLang serving.
inclusionAI Ling 3.1 Flash
inclusionAI Ling 3.1 Flash is available on OpenRouter with 262K context and free pricing.
js-genai v2.27.0
js-genai v2.27.0 adds lyria-3.5, gemini-omni-1.1-flash, and gemini-omni-flash-preview to the Interactions Model enum.
python-genai v2.28.0
Adds lyria-3.5, gemini-omni-1.1-flash, and gemini-omni-flash-preview to the Interactions Model enum and supports continuation_token.
Deepgram Self-Hosted 261001
Self-hosted release 261001 ships Flux TTS watermarking, removes Whisper support, and adds FIPS 140-3 images.
Guided Vision in Gemini Live
Guided Vision launches in Gemini Live for Android.
Pareto 26.10 Preview
Pareto 26.10 Preview is available on OpenRouter at $0.80/$3.20 per 1M tokens with 1M context.
PixelUMM
NVIDIA released PixelUMM, a 15.2B-parameter encoder-free multimodal model for text, image, and video understanding and generation.
java-genai v1.75.0
v1.75.0 adds support for continuation_token in GenerateContent.
js-genai v2.26.0
js-genai v2.26.0 adds support for continuation_token in GenerateContent.
pplx-decider-v1-27b
Perplexity released pplx-decider-v1-27b, a decision model fine-tuned from Qwen3.8-27B, on Hugging Face.
python-genai v2.27.0
python-genai v2.27.0 adds support for continuation_token in GenerateContent.
Anthropic models in Seoul and Singapore
Anthropic models are available for in-region inference in Seoul and Singapore on Amazon Bedrock.
Claude in-country inferencing in India
Claude models are available for in-country inferencing in India on Amazon Bedrock.
Flux TTS Inline Controls
Flux TTS /v2/speak now supports inline pause (500-3000 ms) and pronunciation controls, with pause on batch and pronunciation on batch and streaming.
Gemini 3.8 Flash TTS models
Added Gemini 3.8 Flash TTS and Gemini 3.8 Flash Lite TTS models to the Python SDK.
Google Cloud Lakehouse public datasets
Adds Wikipedia pageviews and GitHub commit histories as public Apache Iceberg datasets in Google Cloud Lakehouse.
java-genai v1.74.0
Adds Gemini 3.8 Flash TTS and Gemini 3.8 Flash Lite TTS models, plus gcs_metrics_uri for tuning job metrics.
js-genai v2.25.0
Adds Gemini 3.8 Flash TTS and Gemini 3.8 Flash Lite TTS models to SDKs.
AdvancedMathBench-AutoVerifier
AutoVerifier model weights, tokenizer, and evaluation code released on Hugging Face.
GPT-6.1 Sol on Amazon Bedrock
GPT-6.1 Sol is now available on Amazon Bedrock.
NVIDIA Kumo Tabular
Open tabular foundation model now on Hugging Face in three sizes from 28M to 215M parameters.
OVIE-512
Kyutai released a 512x512 OVIE checkpoint with 41.6 ms per view on H100.
OVIE-ft-DL3DV
Kyutai released a DL3DV fine-tuned OVIE checkpoint with 0.1B params on Hugging Face.
Claude Sonnet 5.5 on OpenRouter
Claude Sonnet 5.5 is available on OpenRouter at $2/$10 per 1M tokens with 1.0M context.
Eleven v4 and Eleven v4 Turbo
Eleven v4 and Eleven v4 Turbo are available, with Turbo median inference latency of approximately 100 ms.
Grok 4.7 on Amazon Bedrock
Grok 4.7 is now available on Amazon Bedrock.
Mistral Python Client v3.0.0
Python client v3.0.0 released with 24 stable API surface removals.
MiMo-V2.6-Flash-MOPD and MiMo-V2.6-Pro-MOPD
MiMo-V2.6-Flash-MOPD and MiMo-V2.6-Pro-MOPD released on Hugging Face with 1M token context and 309B total parameters.
Health Guardian features
Health Guardian features are rolling out in the Google Health app, with blood pressure trends, insulin resistance trends, and sleep breathing quality metrics included free with Pixel Watch 3, 4, and 5 devices.
Intern-Decision-4B
InternLM releases Intern-Decision-4B, 2B, and 0.8B multimodal structured decision models on Hugging Face.
Perceptron Mk1.5
Perceptron Mk1.5 is available on OpenRouter at $0.15 input and $1.50 output per 1M tokens.
TypeSafe Jev Router
Jev Router is available on OpenRouter with a 1,000,000 token context window and zero pricing.
ChatGPT Ads
ChatGPT Ads now available in more than 60 countries after adding Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, and Taiwan.
Gemini Connected Apps
A new wave of Connected Apps is rolling out to Gemini.
Claude Opus 5.5 on AWS
Claude Opus 5.5 is now available on AWS.
Grok 4.6 in Amazon Bedrock
xAI's Grok 4.6 is now available in Amazon Bedrock.
AgentCore runtime
The new AgentCore runtime delivers a P75 cold start latency of about 2 seconds from a 200 MB image to 2 GB, while the original runtime rose from roughly 5.4 seconds to nearly 30 seconds.
GLM 5.3 FlashX
GLM 5.3 FlashX is available on AI Gateway, serving Z.ai's multimodal coding model at about 200 tokens per second with no markup and no platform fee on inference.
GPT-Live 1
GPT-Live 1 from OpenAI is available on AI Gateway as a full-duplex voice model that can listen and speak at the same time, with client delegation to any text model on AI Gateway.
Claude Fable 5.1
Unchanged at $10 per million input and $50 per million output, with cache reads down to $0.25 per million, and Mythos 5.1 alongside it.
Gemini 3.8 Flash
$0.75 per million input and $3.75 per million output until December 31, 2026, then $1.50 and $7.50 from January 1, 2027.
K2 Horizon
Open weights under Apache 2.0: 375B parameters, 23B active per token, a 512K context window, and a GPQA Diamond score of 87.3 in the model card's own table.
Muse Spark 1.3
Two prices for the same model and the same 1M context window: $0.10 per million input on the contributor tier, where your inputs improve Meta's products, and $1.25 on the standard tier, where they do not.
ChatGPT Epic EHR integration
Healthcare organizations can now connect Epic environments to ChatGPT for Healthcare, and across 4,363 physician ratings, 99.1% of responses were rated safe across 27 clinical use cases.
WeatherNext 3
WeatherNext 3 generates hourly forecasts at 5-kilometer resolution, five times sharper than WeatherNext 2, and is integrated across Google Search, Gemini app, Google Maps, Google Maps Platform Weather API, and Google Earth Engine starting today.
WebGPU kernels
Hugging Face released @huggingface/kernels with 207 WebGPU kernels, 2.57x faster than ORT WebGPU by geometric mean across 809 matching test cases on an Apple M4 GPU.
Claude team plan for scientists
Anthropic is opening 10,000 seats for scientists to access Claude subscriptions free and at discounted rates for one year, with standard seats free and premium seats with 5x usage limits at $15 per month.
GPT-5.6 in Kiro
The GPT-5.6 model family, including Sol, Terra, and Luna, is now available in Kiro, and testing found that on Terminal-Bench 2.1, GPT-5.6 Terra completed successful tasks in Kiro at roughly 82% cost reduction.
ChatGPT for Teachers
OpenAI is expanding ChatGPT for Teachers to 55 additional school systems, bringing it to over 100,000 more educators and staff.
Stable Audio 3.0 plugin
Stability AI ships a DAW plugin for macOS AU and VST3 plus an enhanced StableAudio.com web app, both powered by commercially-safe models where users own outputs and can distribute them freely.
Agentic Search
Mistral Agentic Search improves accuracy up to 3x on FinanceBench (26.7% to 86%) and cuts p90 latency up to 39.6% and token use by one-third.
ChatGPT Ads Europe
ChatGPT Ads expands to 31 European markets, including Germany, France, Spain, Italy, Sweden, Norway, Denmark, the Netherlands, and Austria. Ads show only on Free and Go plans; Plus, Pro, and Enterprise remain ad-free.
GPT-5.6-Cyber
GPT-5.6-Cyber completes 95.0% of advanced cybersecurity requests compared with 1.5% for GPT-5.6 Sol, and is available through Daybreak Red access.
SL2T sign language model
Google DeepMind introduced SL2T, a sign-language-to-text model powering sign-to-text dictation in Gboard and Live Transcribe on Pixel 11, starting with ASL to English.
ChatGPT ads
OpenAI expanded its ChatGPT ads pilot to the UK, Mexico, Brazil, Japan, and South Korea, with ads on Free and Go tiers; Plus, Pro, Business, Enterprise, and Education remain ad-free.
Daybreak on AWS
OpenAI made Daybreak Blue and Daybreak Red available through Amazon Bedrock, giving defenders access to GPT-5.6 Sol and purpose-trained cyber models in AWS environments.
Muse Glimmer 30B
Meta released Muse Glimmer, a 30B multimodal model under Apache 2.0, with day-0 support in transformers, llama.cpp, vLLM, and Inference Endpoints.
Baseten Inference Provider
Baseten is now a supported Inference Provider on the Hugging Face Hub, launching support for conversational and text-generation tasks with open-weight LLMs such as Kimi K3, DeepSeek V4 Flash, and GLM-5.2.
Fable 5 biology safeguards
Anthropic reduced biology-related fallbacks on Claude Fable 5 by about 85% across product surfaces, so users see far fewer switches to Opus 5 on everyday health and educational questions.
ChatGPT for small business program
OpenAI is launching the ChatGPT for small business program with virtual training, in-person AI academies, and new guides, after 78% of participants built a functional AI workflow in a single day at last year's Small Business AI Jams.
Health in ChatGPT
Health in ChatGPT rolls out to U.S. users, letting them connect Apple Health and medical records. GPT-5.6 Sol outperforms GPT-5.5 on HealthBench Professional, and 70% of health conversations happened outside the dedicated experience.
Nunchaku Lite in Diffusers
Nunchaku Lite brings 4-bit diffusion inference to Diffusers, cutting peak VRAM by up to 50% and improving latency by 30%.
Anthropic Economic Index connector
The Anthropic Economic Index connector for Claude lets anyone ask questions about AI usage in the economy, grounded in Index data.
Claude reflection dashboard
Anthropic introduces a beta reflection dashboard in Claude Settings that tracks and visualizes usage patterns over 1, 3, 6, or 12 months, with quiet hours and break nudges.
GPT 5.6 on AI Gateway
GPT 5.6 is available on AI Gateway in limited preview as Sol, Terra, and Luna, with Terra offering performance comparable to the previous generation at half the cost.
GPT-5.6 in Microsoft 365 Copilot
GPT-5.6 becomes the preferred model in Microsoft 365 Copilot across Word, Excel, PowerPoint, Chat, and Cowork, with Microsoft accessing the models directly through the OpenAI API.
vLLM transformers backend
The transformers vLLM backend now meets or beats native throughput on Qwen3 4B dense, 32B dense, and 235B-parameter FP8 MoE models, using torch.fx static analysis and ast source rewriting to apply inference-specific layer fusions at runtime.
Studio prompts and skills
Mistral Studio now provides a system of record for Prompts and Skills, with immutable versions, rollback, clear ownership, classification labels, and audit logs.
Fable 5 redeployed
Claude Fable 5 will be available starting Wednesday, , to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork, with the new safety classifier blocking the reported technique in over 99% of cases.
Gemma 4 voice AI
Hugging Face and Cerebras demonstrate a real-time speech-to-speech pipeline pairing Gemma 4 31B on Cerebras with Nvidia's Parakeet for speech recognition and Alibaba's Qwen3TTS for text-to-speech, already powering more than 9,000 Reachy Mini robots.
Leanstral 1.5
Leanstral 1.5 saturates miniF2F at 100%, solves 587/672 PutnamBench problems, and achieves 87% on FATE-H and 34% on FATE-X, at about $4 per problem against an estimated $300 or more for Seed-Prover 1.5 high.
Mistral OCR 4
Mistral OCR 4 supports 170 languages across 10 language groups, returns bounding boxes, block classification, and inline confidence scores, and runs in a single container for self-hosted deployments, priced at $4 per 1,000 pages with a 50% Batch API discount to $2 per 1,000 pages.
PP-OCRv6
PP-OCRv6 scales from 1.5M to 34.5M parameters across tiny, small, and medium tiers, with the medium and small tiers supporting 50 languages and the medium tier reaching 86.2% detection Hmean and 83.2% recognition accuracy.
Deployment Simulation
OpenAI's Deployment Simulation replays previous conversations with candidate models and achieved a median multiplicative error of 1.5x in predicting deployment-time undesired behavior rates across GPT-5-series Thinking deployments.
LifeSciBench
LifeSciBench includes 750 expert-authored tasks spanning seven workflows and seven biological domains, with 19,020 rubric criteria and 453 expert reviewers.
OpenAI Partner Network
OpenAI is investing $150 million in a partner program and aims to train 300,000 certified consultants by the end of 2026.
Rare disease reanalysis
OpenAI o3 Deep Research reanalyzed 376 previously unsolved rare disease cases and surfaced leads that led to 18 diagnoses, an additional diagnostic yield of 4.8%.
AI Control Roadmap
Google DeepMind's AI Control Roadmap treats AI agents as potential insider threats, using supervisors to monitor and block harmful actions.
UK AI planning prototype
Google DeepMind is partnering with the UK government to build an AI planning prototype aiming to cut application decision times by 50%.
BBVA ChatGPT Enterprise
BBVA now has more than 100,000 employees using ChatGPT Enterprise globally, with 70%+ weekly active usage, around 3 hours saved per employee per week, and up to 80% efficiency gains in selected workflows.
DXC Anthropic alliance
DXC will train tens of thousands of Claude-certified forward-deployed engineers and used Claude to write more than 95% of the code for DXC OASIS, its new AI-native orchestration platform now serving over 50 customers.
Economic Research Exchange
OpenAI launched the OpenAI Economic Research Exchange, a platform for structured project-based collaborations on the economic effects of AI, with applications open until .
Gemma 4 12B
Gemma 4 12B is an encoder-free multimodal model that runs locally with 16GB of VRAM or unified memory, and Gemma 4 models have crossed 150 million downloads.
LSEG OpenAI deployment
LSEG deployed ChatGPT Enterprise and OpenAI APIs across the organization, reducing product release cycles from 3-6 months to 2 weeks and accelerating customer delivery timelines to about 4 weeks from request to production.
Preply Lesson Insights
Preply launched Lesson Insights, an OpenAI API-powered experience that generates tailored feedback across grammar, vocabulary, and pronunciation after each 1:1 lesson, with 75% of English-language learners actively using it and a 4.7/5 satisfaction rating from more than 300k ratings.
TCS Anthropic partnership
TCS will provide Claude to 50,000 of its own employees across 56 countries and build Claude-powered products for clients in financial services, healthcare, the public sector, and other regulated industries.
Claude Fable 5
Claude Fable 5, a Mythos-class model, is generally available at $10 per million input tokens and $50 per million output tokens, with safeguards that trigger in less than 5% of sessions.
Mellum2
JetBrains released Mellum2, a 12B-parameter Mixture-of-Experts model that activates only 2.5B parameters per token and delivers more than 2x faster inference than similar-sized models under the Apache 2.0 license.
Nemotron 3.5 Content Safety
NVIDIA released Nemotron 3.5 Content Safety, a 4B-parameter model that unifies multimodal input, 12-language explicit coverage, custom policy enforcement, and auditable reasoning traces in one inference call.
Project Glasswing expansion
Project Glasswing expands to approximately 150 new organizations, after initial partners found more than 10,000 high- or critical-severity security flaws.
Project Genie Street View
Project Genie now grounds generated worlds in Google Street View imagery for places in the U.S., and access is rolling out to all eligible Google AI Ultra $200 subscribers globally.
OlmoEarth v1.1
OlmoEarth v1.1 is a new family of Earth observation models that cuts compute costs by up to 3x while maintaining OlmoEarth v1's performance on research benchmarks.
Codex on Dell
OpenAI and Dell Technologies partner to bring Codex to hybrid and on-premises environments, with more than 4 million developers now using Codex every week.
PaddleOCR 3.5
PaddleOCR 3.5 brings OCR and document parsing tasks closer to the Hugging Face ecosystem, with supported models able to run with Transformers as an inference backend.
GPT-5.5 on Databricks
Databricks is making GPT-5.5 available through AI Unity Gateway after the model became the first to surpass 50% accuracy on OfficeQA Pro and reduced errors by 46% compared to GPT-5.4.
Gates Foundation partnership
Anthropic and the Gates Foundation committed $200 million in grant funding, Claude usage credits, and technical support over four years for programs in global health, life sciences, education, and economic mobility.
Granite Embedding Multilingual R2
IBM released two Apache 2.0 multilingual embedding models built on ModernBERT: a 97M-parameter compact model scoring 60.3 on MTEB Multilingual Retrieval and a 311M full-size model scoring 65.2, both covering 200+ languages with 32K-token context.
OpenAI Deployment Company
OpenAI is launching the OpenAI Deployment Company with more than $4 billion of initial investment and has agreed to acquire Tomoro, bringing approximately 150 experienced Forward Deployed Engineers and Deployment Specialists from day one.
PwC Claude expansion
PwC will roll out Claude Code and Cowork starting with U.S. teams and expanding toward a global workforce of hundreds of thousands of professionals, with a program to train and certify 30,000 PwC professionals on Claude.
ChatGPT personal finance
ChatGPT personal finance preview lets Pro users in the U.S. connect financial accounts via Plaid, with support for 12,000+ institutions. It provides a dashboard and grounded answers, with Intuit support coming soon.
Codex mobile app
Codex is now in the ChatGPT mobile app in preview, and more than 4 million people use Codex every week.
GPT-5.5-Cyber
GPT-5.5-Cyber is rolling out in limited preview to defenders responsible for securing critical infrastructure, with more permissive behavior for authorized red teaming, penetration testing, and controlled validation, while GPT-5.5 with Trusted Access for Cyber remains the recommended starting point for most security workflows.
ChatGPT ads
OpenAI expands ChatGPT ads with beta self-serve Ads Manager and cost-per-click bidding, adding partners like Dentsu, Omnicom, Publicis, and WPP.
Nemotron 3 Nano Omni
NVIDIA released Nemotron 3 Nano Omni, a 30B-A3B omni-modal model for documents, audio, video, and agentic computer use, with OCRBenchV2-En 65.8, MMLongBench-Doc 57.5, Video-MME 72.2, and VoiceBench 89.4, and checkpoints in BF16, FP8, and NVFP4 on Hugging Face.
Advanced Account Security
OpenAI introduced Advanced Account Security, an opt-in setting requiring passkeys or security keys, disabling email/SMS recovery, and shortening sessions, with Yubico bundle pricing.
DeepInfra on HF
DeepInfra is now a supported Inference Provider on Hugging Face Hub, offering over 100 models including DeepSeek V4, Kimi-K2.6, GLM-5.1, with PRO users getting $2 monthly Inference credits.
Granite 4.1 LLMs
IBM released Granite 4.1, a family of dense LLMs (3B, 8B, 30B) trained on ~15T tokens with 512K context, Apache 2.0, with the 8B matching the previous 32B MoE.
DeepSeek-V4
DeepSeek released V4 today with two MoE checkpoints on the Hub: DeepSeek-V4-Pro at 1.6T total parameters with 49B active, and DeepSeek-V4-Flash at 284B total with 13B active, both with a 1M-token context window.
Agents SDK
The Agents SDK now includes native sandbox execution with built-in support for Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, and Vercel, plus snapshotting and rehydration so losing a sandbox container does not mean losing the run.
Cloudflare Agent Cloud
Cloudflare is expanding access to OpenAI frontier models including GPT-5.4 across Agent Cloud, and the Codex harness is now generally available in Cloudflare Sandboxes.
Codex update
Codex now operates your computer alongside you, works with more than 90 additional plugins, generates images with gpt-image-1.5, and serves more than 3 million developers who use it every week.
Gemini 3.1 Flash TTS
Gemini 3.1 Flash TTS achieved an Elo score of 1,211 on the Artificial Analysis TTS leaderboard and supports 70+ languages with audio tags for controlling vocal style, pace, and delivery.
Multimodal Sentence Transformers
Sentence Transformers v5.4 adds multimodal embedding and reranker models that encode and compare text, images, audio, and video through the same API, with Qwen3-VL-Embedding-2B and Qwen3-VL-Reranker-2B supported.
ChatGPT sales workflows
OpenAI Academy published a guide on using ChatGPT for sales teams, covering account research, meeting prep, follow-ups, deal management, and measuring impact.
OpenAI Axios compromise
OpenAI is revoking and rotating its macOS code signing certificate after the Axios developer tool compromise, requiring users to update macOS apps by .
Falcon Perception
Falcon Perception is a 0.6B-parameter early-fusion Transformer that reaches 68.0 Macro-F1 on SA-Co versus 62.3 for SAM 3, and Falcon OCR is a 0.3B model scoring 80.3 on olmOCR and 88.6 on OmniDocBench.
Granite 4.0 3B Vision
Granite 4.0 3B Vision is available on Hugging Face under Apache 2.0 and leads on PubTablesV2 cropped at 92.1 and full-page at 79.3, OmniDocBench at 64.0, and TableVQA at 88.1.
TRL v1.0
TRL v1.0 is released with a stable core following semantic versioning and an experimental layer, and the library is downloaded 3 million times a month.
mRNA models 25 species
OpenMed trained 4 production mRNA language models across 25 species in 55 GPU-hours for $165, with CodonRoBERTa-large-v2 reaching perplexity 4.10 and Spearman CAI correlation 0.40.
gradio.Server
Gradio released gradio.Server, letting developers pair any custom frontend with Gradio's backend, including queuing and ZeroGPU support.
Spaces CLI
Mistral released Spaces, a CLI built for humans and agents, with every interactive input having a flag equivalent and context files for agent usability.
Lyria 3 Pro
Lyria 3 Pro creates tracks up to 3 minutes long with intros, verses, choruses, and bridges, available in Vertex AI, Google AI Studio, the Gemini API, Google Vids, the Gemini app, and ProducerAI.
Model Spec approach
OpenAI published the backstory of its Model Spec, detailing the Chain of Command, authority levels, decision rubrics, and concrete examples that define intended model behavior.
EVA framework
ServiceNow AI released EVA, an end-to-end evaluation framework for voice agents that jointly scores accuracy (EVA-A) and experience (EVA-X), with an airline dataset of 50 scenarios and results for 20 systems.
Holotron-12B
Holotron-12B, post-trained from NVIDIA Nemotron-Nano-2 VL, raises WebVoyager from 35.1% to 80.5% and achieves over 2x higher throughput than Holo2-8B on a single H100.
Claude Partner Network
Anthropic is committing an initial $100 million to the Claude Partner Network for training courses, dedicated technical support, and joint market development, with membership free of charge and applications open today.
Grok 4.20 on AI Gateway
Grok 4.20 is available on Vercel AI Gateway in Reasoning, Non-Reasoning, and Multi-Agent variants, with the Multi-Agent variant purpose-built for multi-agent orchestration and collaboration.
Promptfoo acquisition
OpenAI is acquiring Promptfoo, an AI security platform trusted by over 25 percent of Fortune 500 companies, and will integrate its technology into OpenAI Frontier for automated security testing and red-teaming.
Prompt injection defenses
OpenAI describes social engineering-based prompt injection attacks and defenses like Safe Url that block silent data transmission.
Responses API computer env
OpenAI's Responses API now includes a shell tool and hosted container workspace for executing real-world tasks with filesystem, databases, and network access.
GPT-5.3 Instant
GPT-5.3 Instant is released as an update to ChatGPT's most-used model, reducing unnecessary refusals and improving web answer synthesis.
Modular Diffusers
Modular Diffusers introduces a new way to build diffusion pipelines by composing reusable blocks, with a quickstart example running FLUX.2 Klein 4B and community pipelines including Krea Realtime Video achieving 11fps on a single B200 GPU.
Codex Security
Codex Security, an application security agent, is in research preview, having scanned 1.2 million commits in 30 days and found 792 critical findings.
OpenAI-Amazon partnership
Amazon will invest $50 billion in OpenAI, starting with $15 billion, and AWS becomes the exclusive third-party cloud distributor for OpenAI Frontier.
Stateful Runtime Environment
OpenAI and Amazon deliver a Stateful Runtime Environment in Amazon Bedrock, powered by OpenAI models, for production agentic workflows.
GGML and llama.cpp join HF
GGML, creators of llama.cpp, are joining Hugging Face, with Georgi Gerganov and team dedicating 100% of their time maintaining llama.cpp with full autonomy and leadership on technical directions and the community.
Lyria 3 music generation
Lyria 3, Google DeepMind's latest generative music model, is rolling out in beta in the Gemini app, letting anyone make 30-second tracks from text or images with custom cover art generated by Nano Banana.
Unsloth and HF Jobs training
Unsloth and Hugging Face Jobs enable fast LLM fine-tuning of LiquidAI/LFM2.5-1.2B-Instruct through coding agents like Claude Code and Codex, with Unsloth providing approximately 2x faster training and about 60% less VRAM usage compared to standard methods.
CUDA kernel agent skill
Hugging Face released an agent skill that teaches Codex and Claude to write production CUDA kernels, producing an RMSNorm kernel for Qwen3-8B with an average 1.94x speedup on H100.
ChatGPT on GenAI.mil
OpenAI deployed a custom ChatGPT on GenAI.mil, the Department of War's secure enterprise AI platform used by 3 million civilian and military personnel, for unclassified work in authorized government cloud infrastructure.
GPT-5.2 physics result
GPT-5.2 Pro conjectured a formula for single-minus gluon tree amplitudes, and an internal scaffolded GPT-5.2 spent roughly 12 hours producing a formal proof, with the preprint available on arXiv.
OpenEnv Calendar Gym
Turing contributed a production-grade calendar management environment to OpenEnv, where agents achieved close to 90% success with explicit calendar identifiers but dropped to roughly 40% with natural language descriptions.
Transformers.js v4
Transformers.js v4 is now available on NPM with a C++ WebGPU runtime, a 10x build time drop from 2 seconds to 200 milliseconds, and a default export that is 53% smaller.
Lockdown Mode
OpenAI introduced Lockdown Mode for ChatGPT Enterprise, Edu, Healthcare, and Teachers, disabling live web access, image support, Deep Research, Agent Mode, and more to reduce prompt-injection risks.
Holo2-235B-A22B Preview
Holo2-235B-A22B Preview achieves 78.5% on Screenspot-Pro in agent mode within 3 steps and 79.0% on OSWorld G, and is available on Hugging Face.
Voxtral Transcribe 2
Voxtral Mini Transcribe V2 is available now via API at $0.003 per minute, and Voxtral Realtime is available via API at $0.006 per minute and as open weights on Hugging Face under Apache 2.0.
Xcode Claude Agent SDK
Xcode 26.3 introduces a native integration with the Claude Agent SDK, giving developers subagents, background tasks, plugins, and visual verification with Xcode Previews inside the IDE.
Mistral Vibe 2.0
Mistral released Mistral Vibe 2.0, a terminal-native coding agent powered by the Devstral 2 model family, with custom subagents, multi-choice clarifications, slash-command skills, and unified agent modes on Le Chat Pro and Team plans.
Project Genie
Google DeepMind rolled out Project Genie to Google AI Ultra subscribers in the U.S. aged 18 and over, a prototype web app powered by Genie 3 that generates interactive worlds in real time from text and image prompts.
ServiceNow chooses Claude
ServiceNow chose Claude as the default model for its ServiceNow Build Agent and as a preferred model across the ServiceNow AI Platform, and is rolling out Claude and Claude Code to its global workforce of more than 29,000 employees.
Daggr
Hugging Face released Daggr, an open-source Python library for building AI workflows that connect Gradio apps, ML models, and custom functions, with automatic visual canvas.
GOV.UK AI assistant
Anthropic partnered with the UK's DSIT to build a Claude-powered AI assistant for GOV.UK, initially focusing on employment support for job seekers.
Horizon 1000
OpenAI and the Gates Foundation are committing $50 million in funding, technology, and technical support to reach 1,000 primary healthcare clinics and their surrounding communities by 2028, beginning in Rwanda.
ServiceNow OpenAI agreement
ServiceNow announced OpenAI will be a preferred intelligence capability for enterprises that run more than 80 billion workflows each year on its platform, with GPT-5.2 built directly into the ServiceNow AI Platform.
AssetOpsBench
IBM Research released AssetOpsBench, a benchmark for industrial AI agents with 2.3M sensor telemetry points and 4.2K work orders.
Diff Transformer V2
Microsoft introduced Differential Transformer V2, improving inference speed and training stability for production LLMs, with experiments still running.
Waypoint-1
Overworld released Waypoint-1, a real-time interactive video diffusion model trained on 10,000 hours of game footage, with weights on the Hub.
GPT 5.2 Codex
GPT 5.2 Codex is now available on Vercel AI Gateway with no other provider accounts required, combining GPT 5.2's professional knowledge work with GPT 5.1 Codex Max's agentic coding.
ChatGPT Go
ChatGPT Go is now available worldwide at $8 per month in the US, offering 10x more messages, uploads, and image creation than the free tier, with GPT-5.2 Instant.
Claude for Healthcare
Anthropic introduced Claude for Healthcare with HIPAA-ready products, connectors to CMS, ICD-10, and NPI, plus new life sciences connectors to Medidata and ClinicalTrials.gov.
Datadog Codex review
Datadog deployed Codex across its engineering workforce, with more than 1,000 engineers using it regularly, and Codex found more than 10 cases, roughly 22% of incidents examined, where its feedback would have made a difference.
Falcon-H1-Arabic
Falcon-H1-Arabic 3B, 7B, and 34B models outperform all SOTA models of similar sizes and sometimes bigger, with the 34B reaching approximately 75% on OALL and surpassing Llama-3.3-70B.
NVIDIA Cosmos Reason 2
NVIDIA released Cosmos Reason 2, an open reasoning vision language model for physical AI that tops the Physical AI Bench and Physical Reasoning leaderboards as the #1 open model for visual understanding.
Netomi agentic systems
Netomi pairs GPT-4.1 for low-latency tool use with GPT-5.2 for multi-step planning, sustaining sub-three-second responses with 98% intent classification accuracy during traffic spikes above 40,000 concurrent customer requests per second.
Tolan voice AI
Tolan built a voice-first AI companion with GPT-5.1, cutting speech initiation time by over 0.7 seconds, dropping memory recall misses by 30%, and raising next-day user retention more than 20%.
Stargate SB Energy partnership
OpenAI and SoftBank each invest $500 million in SB Energy. OpenAI signs 1.2 GW data center lease in Milam County. SB Energy secures $800 million Redeemable Preferred Equity from Ares. Part of $500 billion Stargate commitment.
OpenAI Grove
OpenAI opened applications for Grove, a five-week program for about fifteen pre-idea technical individuals, with mentoring, workshops, and early access to new tools and models.
AprielGuard
ServiceNow released AprielGuard, an 8B parameter safety and security safeguard model that detects 16 safety risk categories and adversarial attacks including prompt injection, jailbreaks, chain-of-thought corruption, context hijacking, memory poisoning, and multi-agent exploit sequences across standalone prompts, multi-turn conversations, and agentic workflows.
ChatGPT Images
The new ChatGPT Images model, powered by GPT Image 1.5, generates images up to 4x faster, makes precise edits while keeping details intact, and image inputs and outputs are now 20% cheaper in GPT Image 1.5 compared to GPT Image 1.
GPT-5.2-Codex System Card
GPT-5.2-Codex is very capable in the cybersecurity domain but does not reach High capability on cybersecurity, and it is being treated as High capability on biology with the corresponding suite of safeguards.
Gemma Scope 2
Gemma Scope 2 is an open suite of interpretability tools for all Gemma 3 model sizes from 270M to 27B parameters, produced by storing approximately 110 Petabytes of data and training over 1 trillion total parameters.
SB 53 compliance framework
Anthropic published its Frontier Compliance Framework for California's SB 53, which goes into effect , covering catastrophic risk assessment and mitigation.
Transformers v5 tokenizers
Transformers v5 redesigns tokenizers, separating architecture from trained vocab, making them inspectable, customizable, and trainable from scratch.
Nemotron 3 Nano evaluation
NVIDIA released the full evaluation recipe for Nemotron 3 Nano 30B A3B using NeMo Evaluator, with scores like 78.3 on MMLU-Pro and 89.1 on AIME 2025.
OpenAI Academy for News
OpenAI launched the Academy for News Organizations with the American Journalism Project and The Lenfest Institute, offering on-demand training and practical use cases.
CUGA on Hugging Face
CUGA, a configurable generalist agent, is now on Hugging Face Spaces, achieving #1 on AppWorld and top-tier on WebArena.
Deutsche Telekom collaboration
Deutsche Telekom will introduce ChatGPT Enterprise across the company and create multilingual AI experiences for its more than 261 million mobile customers, with rollouts beginning in 2026.
Gemini 2.5 Flash Audio
Gemini 2.5 Flash Native Audio leads ComplexFuncBench Audio at 71.5%, with 90% instruction adherence, and adds live speech translation in 70+ languages.
Codex HF skills
Codex can now run end-to-end ML experiments using Hugging Face Skills, including fine-tuning, evaluation, and report generation.
Disney Sora agreement
Disney invests $1 billion in OpenAI and licenses 200+ characters for Sora, with fan videos streaming on Disney+ starting early 2026.
llama.cpp router mode
llama.cpp server now supports router mode, allowing dynamic loading, unloading, and switching between multiple models without restarting.
Claude for Nonprofits
Anthropic launches Claude for Nonprofits with discounts of up to 75% on Team and Enterprise plans, connectors to Blackbaud, Candid, and Benevity, and a free AI Fluency for Nonprofits course.
DeepMath
Intel's DeepMath agent, built on Qwen3-4B Thinking and fine-tuned with GRPO, reduces output lengths by up to 66% while improving accuracy on MATH500, AIME, HMMT, and HLE benchmarks.
DeepSeek V3.2
DeepSeek V3.2 and V3.2 Speciale are now available on Vercel AI Gateway with no other provider accounts required, with V3.2 supporting combined thinking and tool use in both reasoning and non-reasoning modes.
Mistral Large 3
Mistral Large 3 is now available on Vercel AI Gateway with no other provider accounts required, using a sparse mixture-of-experts architecture with 41B active parameters and 675B total.
Nova 2 Lite
Amazon's Nova 2 Lite reasoning model is now available on Vercel AI Gateway with no other provider accounts required, processing text, images, and videos to generate text.
Snowflake partnership
Anthropic and Snowflake announce a multi-year, $200 million agreement making Claude models available to more than 12,600 global customers across Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure, with greater than 90% accuracy on complex text-to-SQL tasks.
Transformers v5
Transformers v5.0.0rc-0 launches with over 400 model architectures, more than 750,000 compatible checkpoints on the Hub, and 1.2 billion total installs, up from 20,000 installs per day at v4.
Trinity Mini
Arcee AI's Trinity Mini, an open weight MoE reasoning model with 26B parameters and 3B active, is now available on Vercel AI Gateway with no other provider accounts required.
swift-huggingface
swift-huggingface is a new Swift package providing a complete client for the Hugging Face Hub with Python-compatible cache.
ChatGPT shopping research
Shopping research in ChatGPT is rolling out today on mobile and web for logged-in users on Free, Go, Plus, and Pro plans, powered by a version of GPT-5 mini trained with reinforcement learning specifically for shopping tasks.
OVHcloud Inference Provider
OVHcloud is now a supported Inference Provider on the Hugging Face Hub, offering serverless access to open-weight models like gpt-oss, Qwen3, DeepSeek R1, and Llama with pay-per-token pricing starting at €0.04 per million tokens.
Claude in Microsoft Foundry
Claude Sonnet 4.5, Haiku 4.5, and Opus 4.1 are now available in public preview in Microsoft Foundry, and Microsoft's Agent Mode in Excel includes an option to use Claude in preview.
Google Antigravity
Google Antigravity is available in public preview at no charge with generous rate limits on Gemini 3 Pro usage, and includes access to Gemini 3, Claude Sonnet 4.5, and OpenAI's GPT-OSS.
Nano Banana Pro
Nano Banana Pro, built on Gemini 3 Pro, generates images with legible text in multiple languages, blends up to 14 images, maintains consistency of up to 5 people, and supports 2K and 4K resolution.
RapidFire AI
Hugging Face TRL now officially integrates with RapidFire AI to accelerate fine-tuning and post-training experiments, with internal benchmarks showing approximately 16-24x higher experimentation throughput than sequential config comparison.
AI image verification in Gemini
Google launched AI image verification in the Gemini app using SynthID, letting users upload an image and ask if it was generated or edited by Google AI.
AnyLanguageModel
AnyLanguageModel is a Swift package that provides a drop-in replacement for Apple's Foundation Models framework, supporting local and remote LLM providers including MLX, llama.cpp, and cloud APIs.
ChatGPT for Teachers
ChatGPT for Teachers is free for verified U.S. K-12 educators through June 2028, with education-grade security and admin controls.
ROCm kernel builder
Hugging Face's kernels library now supports building and sharing ROCm kernels, with a guide using the RadeonFlow GEMM kernel for MI300X.
AI-orchestrated espionage
Anthropic reports a Chinese state-sponsored group manipulated Claude Code to attempt infiltration into roughly thirty global targets, with AI performing 80 to 90% of the campaign.
Free ChatGPT for veterans
U.S. servicemembers and veterans within 12 months of retirement or separation can receive a free year of ChatGPT Plus.
Group chats in ChatGPT
Group chats roll out to all logged-in users on ChatGPT Free, Go, Plus and Pro plans globally over the coming days, with responses powered by GPT-5.1 Auto.
Maryland Claude partnership
Maryland will deploy Claude across multiple state agencies to connect families with benefits and help caseworkers verify more than 150,000 documents each month.
NYT privacy fight
OpenAI says the New York Times is demanding 20 million private ChatGPT conversations and that OpenAI is fighting the demand while developing client-side encryption.
OpenAI for Ireland
OpenAI launches OpenAI for Ireland with an SME Booster programme in 2026, a Dogpatch Labs partnership, and a three-year Patch partnership for founders aged 16 to 21.
Philips AI literacy
Philips is scaling AI literacy across 70,000 employees with ChatGPT Enterprise, moving from individual productivity to workflow-level automation.
Political even-handedness
Anthropic open-sources an automated evaluation showing Claude Sonnet 4.5 scores 94% on political even-handedness, compared with GPT-5 at 89% and Llama 4 at 66%.
SIMA 2
SIMA 2 integrates Gemini models to reason about goals, converse with users, and improve itself through self-directed play in virtual 3D worlds.
GPT-5.1
GPT-5.1 Instant and Thinking roll out to all users, with warmer tone, adaptive reasoning, and faster simple tasks. API release includes no-reasoning mode, 24-hour prompt caching, and new apply_patch and shell tools.
1 million business customers
OpenAI announced more than 1 million business customers around the world are directly using OpenAI, with more than 7 million total ChatGPT for Work seats, up 40% in just 2 months.
CRED Cleo
CRED built Cleo, an AI conversational companion powered by OpenAI models including GPT-4.0, GPT-5, and o3, and in three months since launch Cleo reached a 98% resolution accuracy rate with a 14 percentage point improvement in CSAT scores.
IndQA benchmark
OpenAI released IndQA, a new benchmark that evaluates AI systems on Indian culture and languages, spanning 2,278 questions across 12 languages and 10 cultural domains, created in partnership with 261 domain experts.
Cognizant Claude deployment
Cognizant will deploy Claude to up to 350,000 employees globally, combining Claude with agentic tooling, engineering platforms, and industry blueprints to accelerate enterprise AI adoption and internal transformation.
Aardvark
OpenAI announced Aardvark, an agentic security researcher powered by GPT-5 that identified 92% of known and synthetically-introduced vulnerabilities in benchmark testing and is now in private beta.
Granite 4.0 Nano
IBM released Granite 4.0 Nano models from 350M to 1.5B parameters under Apache 2.0, trained on over 15T tokens with native support on vLLM, llama.cpp, and MLX.
Streaming datasets
Hugging Face improved streaming datasets with 100x fewer startup requests, 10x faster data resolution, and up to 2x faster streaming speed, outrunning local SSDs when training on 64xH100 with 256 workers.
huggingface_hub v1.0
huggingface_hub reached v1.0 with 113.5 million monthly downloads, powering access to over 2 million public models, 0.5 million public datasets, and 1 million public Spaces.
gpt-oss-safeguard
OpenAI released gpt-oss-safeguard-120b and 20b, open-weight reasoning models for safety classification under Apache 2.0, downloadable from Hugging Face.
ChatGPT sensitive conversations
OpenAI updated ChatGPT's default model to reduce undesired responses in sensitive conversations by 65-80%, working with over 170 mental health experts.
Claude for Life Sciences
Claude Sonnet 4.5 scores 0.83 on Protocol QA against a human baseline of 0.79, and Anthropic is adding connectors to Benchling, BioRender, PubMed, Scholar Gateway, Synapse.org, and 10x Genomics plus a single-cell-rna-qc Agent Skill.
LeRobot v0.4.0
LeRobot v0.4.0 introduces Datasets v3.0 with chunked episodes and streaming, integrates PI0, PI0.5, and GR00T N1.5 policies, adds LIBERO and Meta-World simulation support, and launches a plugin system for third-party hardware.
MedGemma 27B Multimodal
MedGemma 27B Multimodal adds image and text inputs, scoring 87.7% on MedQA, within 3 points of DeepSeek R1 at one tenth the inference cost.
Sentence Transformers
Sentence Transformers is transitioning from the UKP Lab at TU Darmstadt to Hugging Face, with over 16,000 models on the Hub serving more than a million monthly unique users, and Tom Aarsen continuing as maintainer.
UK data residency
OpenAI is introducing UK data residency on Friday 24th October for API Platform, ChatGPT Enterprise, and ChatGPT Edu customers, and a new agreement with the UK Ministry of Justice will provide 2,500 employees with ChatGPT Enterprise access.
VirusTotal scanning
Every one of the 2.2M+ public model and dataset repositories on the Hugging Face Hub is being continuously scanned with VirusTotal, comparing file hashes against its threat-intelligence database without sharing raw file contents.
AI for Food Allergies
Hugging Face released Awesome Food Allergy Datasets, the first open collection of datasets on food allergies, to accelerate AI research.
Agent Skills
Agent Skills are folders with instructions, scripts, and resources that Claude loads when relevant, available across Claude apps, Claude Code, and API, with a new /v1/skills endpoint for versioning.
GPT OSS on Intel Xeon
Intel and Hugging Face benchmarked GPT OSS on Google C4 VMs with Intel Xeon 6, finding a 1.7x TCO improvement over C3 VMs.
Plex Coffee with ChatGPT
Plex Coffee cut onboarding time from weeks to days and reduced operational WhatsApp questions by over 50% using ChatGPT Business.
Expert Council on Well-Being
OpenAI assembled the Expert Council on Well-Being and AI, with members from psychology and psychiatry, to guide ChatGPT and Sora development.
Stargate Argentina
OpenAI and Sur Energy signed a Letter of Intent to explore a large-scale data center project in Argentina, potentially the first Stargate in Latin America.
Deloitte Claude alliance
Anthropic and Deloitte expanded their alliance to make Claude available to more than 470,000 Deloitte people and to train and certify 15,000 professionals on Claude.
Codex
Codex is generally available with Slack integration, SDK, and admin tools. Daily usage grew 10x since early August; GPT-5-Codex served over 40 trillion tokens in three weeks. Cisco reports 50% faster code reviews.
Qwen3-8B agent speedup
OpenVINO.GenAI accelerates Qwen3-8B generation by about 1.4x on Intel Core Ultra using speculative decoding with a depth-pruned Qwen3-0.6B draft model.
RTEB benchmark
RTEB is a new retrieval embedding benchmark in beta that combines open and private datasets across 20 languages and domains like law, healthcare, code, and finance, using NDCG@10 as the default metric.
Sora 2 System Card
Sora 2 is available via sora.com, in a new standalone iOS Sora app, and in the future it will be available via the API, with initial access rolling out through limited invitations.
dots.ocr on-device
Hugging Face converted dots.ocr, a 3B parameter model that surpasses Gemini 2.5 Pro on OmniDocBench, to Core ML and MLX, but the initial conversion is over 5GB and slow.
GDPval
GDPval evaluates models on 1,320 tasks across 44 occupations, finding frontier models approach expert quality and complete tasks 100x faster and cheaper.
Shared projects and connectors
Shared projects for Business, Enterprise, Edu plans; new connectors to Gmail, Calendar, Outlook, Teams, SharePoint, GitHub, Dropbox, Box; ISO 27001/27017/27018/27701 certifications; SOC 2 expanded.
LeRobotDataset v3.0
LeRobotDataset v3.0 packs multiple episodes in a single file using relational metadata, and natively supports streaming mode to process large datasets on the fly without downloading prohibitively large collections onto disk.
Scaleway Inference Provider
Scaleway is now a supported Inference Provider on the Hugging Face Hub, offering serverless access to models like gpt-oss, Qwen3, DeepSeek R1, and Gemma 3 with pay-per-token pricing starting at €0.20 per million tokens.
Image Services on Bedrock
Stability AI launched Image Services on Amazon Bedrock, offering nine editing tools like Inpaint, Erase, and Remove Background as managed API services for enterprise workflows.
Public AI Inference Provider
Public AI is now a supported Inference Provider on Hugging Face, offering free access to public and sovereign models like Apertus-70B, with usage free of charge at the time of writing.
One line every time AI actually ships, around the clock, free on Threads and Bluesky. Every one is in The Deploy Log with its source.