OpenAI's Decisions API goes GA, Recursive releases Flint, Agent Skills moves to the AAIF, 1Password launches agent sign-in

The Decisions API (POST /v1/decisions) answers questions you define in a fixed format instead of writing free text. There are three question types: predicate returns the probability that something is true, choice picks one of your options, and score rates input on ordered levels. It runs on gpt-6-luna and takes text, images and JSON. With GA, the limit per request doubled to 400 questions, and OpenAI added prompt caching, a GPT-Live integration and US and EU processing regions.

curl https://api.openai.com/v1/decisions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-6-luna", "input": "I was charged twice for my order.", "questions": [{"type": "choice", "name": "department", "instructions": "Which department should handle this complaint?", "choices": [{"value": "billing", "description": "Payments, invoices, and refunds."}, {"value": "technical", "description": "Problems using the product."}, {"value": "shipping", "description": "Delivery and tracking."}]}]}'

Agents make lots of small calls inside one task: which tool to use, where a ticket should go. Running those through a general response API adds latency and cost. OpenAI says Decisions is up to 10x faster than calling the same model through the Responses APIVendor claim. Uncached input costs $0.10 per million tokens and output tokens are free. If you have steps that only need a judgment, like ticket triage, content moderation or agent routing, it's worth pricing them out on this API.

Flint is the first model from Recursive, the company Richard Socher founded. It's a mixture-of-experts (MoE) model with 40B total parameters, of which only 2.5B are active per token, so it runs fast on a laptop. Context goes up to 512k tokens, and the company recommends staying at 256k or below. The weights are on Hugging Face under Apache 2.0 and work with llama.cpp, Ollama, LM Studio, vLLM and SGLang.

hf download RecursiveSI/flint

Recursive reports more than 125 tokens per second on an M4 Pro MacBook ProVendor claim and more than 150 on an RTX SparkVendor claim. It also says Flint leads MoE models under 50B on SWE-Bench Pro v2, Terminal-Bench 2.1 and SWE-Bench VerifiedVendor claim, and that the NVFP4 quantized model lands within 0.7 points of the 16-bit oneVendor claim. Teams that can't send code to an outside API can try it as a local coding agent. The company says much of the model was built by its own automated research system, the Eureka Machine.

Agent Skills is a standard for teaching an agent a specific task with one folder: a SKILL.md file plus any scripts and reference material it needs. Anthropic published it in December 2025 and has now handed it to the Agentic AI Foundation (AAIF) under the Linux Foundation, where MCP already lives. The name and logo, the GitHub organization and the agentskills.io domain all move over. Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals, with David Soria Parra and Mahesh Murag as lead maintainers.

Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, support the standard, and that 1 million skills have been published with nearly 280 million installsVendor claim. With a neutral foundation in charge instead of one vendor, a team's internal procedures written as skills are more likely to keep working across agents. A roadmap for versioned releases, a formal proposal process and security guidance for writing and installing skills is coming, so teams that ship skills will need to track spec changes.

1Password for Agents fills in your login for a personal AI agent instead of handing the agent the password. The agent proves who it is with a unique digital key, and you see which credential it needs and why, then approve on your own device. 1Password fills the sign-in form directly, and the agent carries on without ever seeing the password. The first integrations are Meta's Muse, Claude Duet, ChatGPT Work and dots, Grok Bot and Town.

Paste a password into an agent and it can end up in chat history or memory. Here, each sign-in needs its own approval and the credential stays outside the model, which cuts that risk. Browserbase, Browser Use, Perplexity, Ramp and others are also listed as partners, so if you build browser agents, check this integration before writing your own login handling.