The Deploy Log | Month by month
47 rows dated December 2025 are in The Deploy Log: 9 leads, 23 in Models and APIs, 14 in Tools and products and 1 in Robotics, hardware and chips, from 8 publishers. They include AI SDK 6 from Vercel, GLM-4.7 on AI Gateway from Vercel, MiniMax M2.1 on AI Gateway from Vercel, Gemini 3 Flash from Google DeepMind, BBVA ChatGPT Enterprise from OpenAI and 42 others. Every item sourced to the publisher's own page.
AI SDK 6 adds an Agent abstraction, ToolLoopAgent with a 20-step default loop, tool execution approval via a needsApproval flag, strict mode, MCP tool calling with structured output, DevTools, reranking, standard JSON Schema, and image editing. Thomson Reuters built CoCounsel with 3 developers in 2 months and now serves 1,300 accounting firms.
Z.ai GLM-4.7 is live on Vercel AI Gateway with no other provider accounts required. It brings major improvements in coding, tool usage, multi-step reasoning, complex agentic tasks, conversational tone, and front-end aesthetics.
MiniMax M2.1 is live on Vercel AI Gateway with no other provider accounts required. It is faster than M2 with improvements in coding, multi-step tool call tasks, refactoring, feature adds, bug fixes, and code review across Go, C++, JS, C#, and TS.
Gemini 3 Flash offers Pro-grade reasoning at Flash-level latency, efficiency, and cost. It scores 90.4% on GPQA Diamond, 33.7% on Humanity's Last Exam without tools, 81.2% on MMMU Pro, and 78% on SWE-bench Verified, outperforming Gemini 2.5 Pro and Gemini 3 Pro on that coding benchmark. It uses 30% fewer tokens on average than 2.5 Pro and is 3x faster.
BBVA expands ChatGPT Enterprise to all 120,000 global employees across 25 countries, a 10x increase from 11,000. Employees save nearly three hours per week on routine tasks and more than 80% engage daily.
GPT-5.2 Thinking scores 70.9% on GDPval knowledge work tasks, 55.6% on SWE-Bench Pro, 80.0% on SWE-bench Verified, 92.4% on GPQA Diamond, and 98.7% on Tau2-bench Telecom. GPT-5.2 Instant, Thinking, and Pro are available now in the API to all developers.
Claude Code reached $1 billion in run-rate revenue in November, six months after becoming generally available in May 2025. Anthropic is acquiring Bun, the JavaScript runtime with more than 7 million monthly downloads and over 82,000 GitHub stars.
OpenAI's GPT-5.1 Codex Max is now available on Vercel AI Gateway with no other provider accounts required. The model uses compaction to operate across multiple context windows and is optimized for long-running coding tasks.
Mistral 3 ships three dense models at 14B, 8B, and 3B parameters plus Mistral Large 3, a sparse mixture-of-experts with 41B active and 675B total parameters, all under Apache 2.0. The 14B reasoning variant hits 85% on AIME '25.
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.
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 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 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.
Anthropic published its Frontier Compliance Framework for California's SB 53, which goes into effect January 1, covering catastrophic risk assessment and mitigation.
Transformers v5 redesigns tokenizers, separating architecture from trained vocab, making them inspectable, customizable, and trainable from scratch.
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 launched the Academy for News Organizations with the American Journalism Project and The Lenfest Institute, offering on-demand training and practical use cases.
CUGA, a configurable generalist agent, is now on Hugging Face Spaces, achieving #1 on AppWorld and top-tier on WebArena.
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 Native Audio leads ComplexFuncBench Audio at 71.5%, with 90% instruction adherence, and adds live speech translation in 70+ languages.
Codex can now run end-to-end ML experiments using Hugging Face Skills, including fine-tuning, evaluation, and report generation.
Disney invests $1 billion in OpenAI and licenses 200+ characters for Sora, with fan videos streaming on Disney+ starting early 2026.
llama.cpp server now supports router mode, allowing dynamic loading, unloading, and switching between multiple models without restarting.
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.
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 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 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.
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.
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 v5.0.0rc-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.
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 is a new Swift package providing a complete client for the Hugging Face Hub with Python-compatible cache.
Vercel released a v0 template that turns a selfie into a 16-bit pixel trading card using GPT Image 1.5 Edit, BRIA RMBG 2.0 background removal, Canvas API compositing, and a custom LoRA trained on pixel art, with the on-site pipeline completing in roughly six seconds.
Vercel's v0 AI agent lets business users build and publish internal tools with auditable code, enforced access control, and secure-by-default infrastructure.
Vercel's bulk redirects are now generally available via UI, API, or CLI without a new deployment, allowing up to one million static URL redirects per project, with Pro plans including 1,000 redirects and Enterprise including 10,000.
Cline now runs on the Vercel AI Gateway, and over a week of live A/B testing, P99 streaming latency improved by 10-14% across Cline's most-used models while API error rates dropped by 43.8%.
Vercel now supports dynamic URL prefixes for multi-tenant platforms, allowing a single project to be backed by tens of thousands of domains with prefixes like tenant-123---project-name-git-branch.yourdomain.dev.
Vercel Agent in the Dashboard can now interact with installed Marketplace integrations from Neon, Supabase, Dash0, Stripe, Prisma, and Mux, with tools exposed by providers available automatically and authentication handled by Vercel.
Gemini 3 Flash is now available on Vercel's AI Gateway, beating Gemini 2.5 Pro across most benchmarks while using 30% less tokens and being 3x faster.
Vercel AI Gateway now serves GPT-5.2 Chat, GPT-5.2, and GPT-5.2 Pro with no other provider accounts required.
Vercel Sandbox now supports Node.js version 24 via @vercel/sandbox 1.1.0 or above with the runtime property set to node24.
Vercel Agent now detects vulnerable React, Next.js, and related RSC packages and automatically generates pull requests with verified fixes for React2Shell CVE-2025-55182 at no cost.
Vercel launches first-class Rust runtime support in public beta with Fluid compute, HTTP response streaming, Active CPU pricing, and an environment variable limit raised from 6KB to 64KB.
Vercel Web Analytics now supports splitting data across 11 dimensions including paths, routes, host names, countries, devices, OS, referrers, flags, flag values, event names, and event properties.
Vercel releases npx fix-react2shell-next to update affected Next.js apps, with Vercel Agent able to open pull requests that upgrade vulnerable projects and WAF rules filtering known exploit patterns.
Notion Agent can now build forms in seconds, collecting requests, feedback, and surveys without manual form building.
Waymo served over 14 million trips in 2025, tripling public rides, and achieved a 10-fold reduction in serious injury crashes compared to human drivers.
The edition every row came from arrives in your inbox, free.
Join free