Access becomes a request, not a stored secret
ChatGPT Enterprise spend controls, GPT-5.5 Instant health and GLM-5.2, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
OpenAI
ChatGPT Enterprise spend controls
DEPLOYif you administer ChatGPT Enterprise
WHAT SHIPPEDCredit usage analytics and updated spend controls are available today in the Global Admin Console. Admins can track usage trends, identify top users, break down credit spend by user, product, and model, set default workspace limits, configure group limits, and create individual overrides.
THE CATCHThe analytics unify ChatGPT and Codex credit usage in one view, and the same data is available through a unified Cost API for deeper analysis in your own systems.
IF YOU RUNa ChatGPT Enterprise workspace with multiple teams and unpredictable spend
SOURCEOpenAI, ChatGPT Enterprise spend controls
OpenAI
GPT-5.5 Instant health
DEPLOYif you answer health questions in ChatGPT
WHAT SHIPPEDGPT-5.5 Instant now performs at a level comparable to frontier Thinking models on OpenAI's most challenging health evaluations. The rate of responses with at least one flagged factuality issue in health production traffic has fallen by 71% in the last two months.
THE CATCHPhysicians rated GPT-5.5 Instant responses higher than physician-written responses across criteria in a 3,500-response evaluation, including fewer instances of missing red flags or failing to seek additional context.
IF YOU RUNa consumer health or wellness product that relies on ChatGPT
SOURCEOpenAI, Improving health intelligence in ChatGPT
Hugging Face
GLM-5.2
DEPLOYif you run long-horizon coding agents
WHAT SHIPPEDGLM-5.2 is an MIT-licensed open-source model with a 1M-token context. It scores 81.0 on Terminal-Bench 2.1 and 62.1 on SWE-bench Pro, and is the highest-ranked open-source model on FrontierSWE, PostTrainBench, and SWE-Marathon.
THE CATCHIndexShare reuses one indexer across every four sparse attention layers, cutting per-token FLOPs by 2.9x at 1M context, and the MTP layer gains up to 20% acceptance length for speculative decoding.
IF YOU RUNcoding agents that need to sustain quality across long, messy trajectories
SOURCEHugging Face, GLM-5.2: Built for Long-Horizon Tasks
The Deploy Log
Models and APIs
| WATCHif you sell enterprise AI services 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. OpenAI |
| WATCHif you work in clinical genomics 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%. OpenAI |
| WATCHif you evaluate life science models LifeSciBench. LifeSciBench includes 750 expert-authored tasks spanning seven workflows and seven biological domains, with 19,020 rubric criteria and 453 expert reviewers. OpenAI |
| WATCHif you run pre-deployment safety reviews 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. OpenAI |
Tools and products
| DEPLOYif you run long agentic workloads Vercel Sandbox 24 hours. Vercel Sandboxes can now run uninterrupted sessions for up to 24 hours, up from 5 hours, on all Pro and Enterprise plans. Vercel |
| DEPLOYif you run long LLM reasoning calls Vercel Functions 30 minutes. Vercel Functions on Node.js and Python runtimes now support execution durations up to 30 minutes for Pro and Enterprise teams, more than 2x the previous 800 second limit. Vercel |
| DEPLOYif you run durable workflows with timeouts Workflow SDK cancellation. Workflow SDK 5 beta supports AbortController and AbortSignal across workflow and step boundaries, with signals staying durable across suspensions and deterministic replay. Vercel |
| DEPLOYif you secure agentic workflows Auth0 on Vercel Marketplace. Auth0 is now available on the Vercel Marketplace with automatic provisioning, Next.js SDK support, and configuration sync across Development, Preview, and Production environments. Vercel |
| DEPLOYif you organize data by project Blob store Hobby limit. Hobby users can now create up to 100 Blob stores, up from 5, with storage, operations, and transfer limits still applying. Vercel |
Robotics, hardware and chips
| DEPLOYif you train robot policies from demonstrations Strands Robots. Strands Robots is an Apache 2.0 SDK from AWS that exposes robot abstractions, simulation, and the LeRobot stack as AgentTools, with sim and hardware datasets sharing the same on-disk format. Hugging Face |
Breakthroughs
| WATCHif you build robot manipulation policies MolmoMotion. MolmoMotion forecasts 3D point trajectories from video, instructions, and query points, and a control policy built on it succeeds on 76.3% of pick-and-place tasks versus 56.0% for Molmo 2. Allen Institute for AI |
Left out
- DAAAM: a figure in the draft could not be found on the publisher's own page
- GLM 5.2 on AI Gateway: the publisher's page does not show it available yet
- Chat SDK Telegram rich text: the publisher's page does not show it available yet
- AI chemist Chan-Lam: the publisher's page does not show it available yet
- Agentic Resource Discovery: it did not carry a number that moved
- Workflow SDK TanStack Start: it did not carry a number that moved
- Alloy modeling: it did not carry a number that moved
- CLI deployment limits: it did not carry a number that moved
- Vercel Passport: it did not carry a number that moved
- eve: it did not carry a number that moved
- Vercel Connect: it did not carry a number that moved
The pattern
The week splits into two regimes. OpenAI ships spend controls, GPT-5.5 Instant health gains, a $150 million partner program, 300,000 certified consultants by 2026, rare disease reanalysis, LifeSciBench, and Deployment Simulation. Vercel ships 24 hour Sandboxes, 30 minute Functions, Workflow SDK 5 cancellation, Auth0 on the Marketplace, and 100 Blob stores for Hobby users. Hugging Face ships GLM-5.2, an MIT licensed 1M token context model, and Strands Robots, an Apache 2.0 SDK from AWS. Allen Institute for AI ships MolmoMotion. The first regime is access as a request, not a stored secret: spend controls, partner certification, and Deployment Simulation all make model behavior and cost legible before and during use, while GPT-5.5 Instant cuts flagged factuality issues by 71% and physicians rate it above physician written responses across 3,500 responses. The second regime is duration and reach: Vercel extends Sandboxes from 5 hours to 24 hours and Functions from 800 seconds to 30 minutes, GLM-5.2 sustains quality across long trajectories with IndexShare cutting per token FLOPs by 2.9x at 1M context, and MolmoMotion forecasts 3D point trajectories for 76.3% pick and place success versus 56.0% for Molmo 2.
Together these items say that the bottleneck has moved from model capability to operational control and long horizon execution. OpenAI's Deployment Simulation predicts undesired behavior with a median multiplicative error of 1.5x across GPT-5 series Thinking deployments, and its rare disease reanalysis of 376 cases yields 18 diagnoses, a 4.8% additional diagnostic yield. Vercel's Workflow SDK 5 beta makes AbortController and AbortSignal durable across suspensions and deterministic replay, while Auth0 automatic provisioning syncs Development, Preview, and Production. Hugging Face's Strands Robots exposes robot abstractions, simulation, and the LeRobot stack as AgentTools with sim and hardware datasets sharing the same on disk format. No single item says that access, spend, duration, and cancellation are now the product surface, but together they show a stack where a 1M token context model, a 24 hour sandbox, a 30 minute function, and a 4.8% diagnostic yield all require the same thing: control that survives the length of the task.