Releases push capability forward as deployment layer races to catch up
GPT-5.6, GPT-Live and ChatGPT Work, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
OpenAI
GPT-5.6
DEPLOYif you run agentic coding or knowledge work
WHAT SHIPPEDGPT-5.6 family launches in general availability with Sol, Terra, and Luna. Sol scores 53.6 on Agents' Last Exam, 80 on the Artificial Analysis Coding Agent Index, 92.2% on BrowseComp, and 62.6% on OSWorld 2.0. Terra and Luna outperform Fable 5 at around one-sixteenth the cost.
THE CATCHThe efficiency numbers are the real story: Sol beats Fable 5 by 11.4 points at medium reasoning at roughly one-quarter the estimated cost, and Terra and Luna outperform Fable 5 at around one-sixteenth the cost. The ultra setting coordinates four agents in parallel by default.
IF YOU RUNa team paying per token for frontier models, the cost-per-task drop is the number to watch
SOURCEOpenAI, GPT-5.6: Frontier intelligence that scales with your ambition
OpenAI
GPT-Live
DEPLOYif you build voice interfaces
WHAT SHIPPEDGPT-Live launches with GPT-Live-1 and GPT-Live-1 mini rolling out to ChatGPT users globally on iOS, Android, and ChatGPT.com. It uses a full-duplex architecture that listens and speaks at the same time, and delegates deeper work to GPT-5.5 in the background.
THE CATCHThe delegation architecture decouples continuous interaction from reasoning, so GPT-Live can keep talking while a frontier model handles search or complex work in the background. Each week more than 150 million people already talk to ChatGPT using Voice and Dictation.
IF YOU RUNa support or sales voice agent, the full-duplex turn-taking is the upgrade to test
SOURCEOpenAI, Introducing GPT-Live
OpenAI
ChatGPT Work
DEPLOYif you run multi-step office workflows
WHAT SHIPPEDChatGPT Work launches on web and mobile for Pro, Enterprise, and Edu plans, with Plus and Business following over the next few days. It is powered by GPT-5.6 and can gather information across apps, create sheets, slides, docs, and web apps, and stay with projects for hours.
THE CATCHScheduled Tasks let ChatGPT Work keep projects moving when you are away, turning new Teams and Slack messages into updated docs or slides. Nearly 100% of teams inside OpenAI now use ChatGPT Work and Codex, and finance reduced month-end close from days to hours.
IF YOU RUNa team where repetitive doc and slide updates eat analyst time, this is the week to pilot it
SOURCEOpenAI, ChatGPT is now a partner for your most ambitious work
The Deploy Log
Models and APIs
| DEPLOYif your org pays for Microsoft 365 Copilot 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. OpenAI |
| DEPLOYif you route through AI Gateway 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. Vercel |
| WATCHif you manage AI usage policy 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. Anthropic |
| DEPLOYif you serve Hugging Face models with vLLM 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. Hugging Face |
Tools and products
| DEPLOYif you evaluate many feature flags per request Flags SDK 10x faster. Flags SDK and Vercel Flags now evaluate multiple feature flags in bulk around 10x faster by reducing microtask queue overhead and creating fewer promises. Vercel |
| WATCHif you build agentic infrastructure Vercel acquires Better Auth. Vercel acquires Better Auth, the open source TypeScript authentication library with 4.7M+ weekly npm downloads and more than 850 contributors, with the team continuing work on Agent Auth for scoped, revocable agent identity. Vercel |
Robotics, hardware and chips
| WATCHif you run warehouse or delivery robots Robostral Navigate. Robostral Navigate is an 8B model that navigates robots with a single RGB camera, achieving 76.6% success on unseen R2R-CE benchmarks and beating the best depth or multi-camera system by 4.5 points. Mistral |
Breakthroughs
| WATCHif you plan around weather-dependent energy or transport Aurora 1.5. Aurora 1.5 adds 22 new weather variables, hourly temporal resolution, and probabilistic ensemble forecasting, outperforming ECMWF ENS on 88.9% of evaluated variable-and-lead-time targets. Microsoft Research |
| DEPLOYif you train robot policies LeRobot v0.6.0. LeRobot v0.6.0 ships world model policies VLA-JEPA, FastWAM, and LingBot-VA, new VLAs including GR00T N1.7 and MolmoAct2, reward models Robometer and TOPReward, and six new simulation benchmarks under lerobot-eval. Hugging Face |
| DEPLOYif you build robot manipulation systems MolmoAct 2. MolmoAct 2, Ai2's vision-language-action model, is available with model weights, data, training code, and fine-tuning scripts, and won an embodied AI hackathon with a voice-controlled robot ten days after release. Allen Institute for AI |
Left out
- Seedream 5.0 Pro on AI Gateway: the publisher's page could not be read when this edition was built
- Grok 4.5 on AI Gateway: the publisher's page does not show it available yet
- Muse Spark 1.1 on AI Gateway: the publisher's page does not show it available yet
- FloatForm robot boats: the publisher's page does not show it available yet
- SkyPilot zero-egress storage: it did not carry a number that moved
- Hugging Face models on Foundry: it did not carry a number that moved
- One-click SageMaker Studio: it did not carry a number that moved
- Notion Workers sharing: it did not carry a number that moved
- Traces Tree and Waterfall views: it did not carry a number that moved
- Studio prompts and skills: it did not carry a number that moved
- Lovable apps on Vercel: the condition it attaches to the call did not hold up
- Notion Agents iOS app: it did not carry a number that moved
The pattern
GPT-5.6 family launches in general availability with Sol, Terra, and Luna, and the efficiency numbers are the real story: Sol beats Fable 5 by 11.4 points at medium reasoning at roughly one-quarter the estimated cost, and Terra and Luna outperform Fable 5 at around one-sixteenth the cost. The same week, GPT-Live launches with GPT-Live-1 and GPT-Live-1 mini rolling out to ChatGPT users globally on iOS, Android, and ChatGPT.com, using a full-duplex architecture that listens and speaks at the same time, and ChatGPT Work launches on web and mobile for Pro, Enterprise, and Edu plans, powered by GPT-5.6, with Scheduled Tasks that keep projects moving when you are away. GPT-5.6 becomes the preferred model in Microsoft 365 Copilot across Word, Excel, PowerPoint, Chat, and Cowork, and 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.
The deployment layer races to catch up: Vercel acquires Better Auth, the open source TypeScript authentication library with 4.7M+ weekly npm downloads and more than 850 contributors, with the team continuing work on Agent Auth for scoped, revocable agent identity, while Flags SDK and Vercel Flags now evaluate multiple feature flags in bulk around 10x faster. Hugging Face ships LeRobot v0.6.0 with world model policies VLA-JEPA, FastWAM, and LingBot-VA, new VLAs including GR00T N1.7 and MolmoAct2, reward models Robometer and TOPReward, and six new simulation benchmarks under lerobot-eval, and the transformers vLLM backend now meets or beats native throughput on Qwen3 4B dense, 32B dense, and 235B-parameter FP8 MoE models. Mistral introduces Robostral Navigate, an 8B model that navigates robots with a single RGB camera, achieving 76.6% success on unseen R2R-CE benchmarks, and Ai2 releases MolmoAct 2 with model weights, data, training code, and fine-tuning scripts, winning an embodied AI hackathon with a voice-controlled robot ten days after release. Microsoft Research adds Aurora 1.5 with 22 new weather variables, hourly temporal resolution, and probabilistic ensemble forecasting, outperforming ECMWF ENS on 88.9% of evaluated variable-and-lead-time targets, while 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.