AI Agents Weekly：GPT-5.6、Ornith-1.0、Codex Inside OpenAI、Claude Tag、Qwen-AgentWorld、AI SDK 7 等

In today's issue:

• OpenAI previews the GPT-5.6 family

• Ornith-1.0 ships open coding models

• OpenAI: agents reshape every department

• Claude Tag joins your Slack team

• Qwen open-sources AgentWorld world model

• Cursor exposes benchmark reward hacking

• Vercel ships AI SDK 7

• OpenRouter MCP picks your model

• Mistral launches OCR 4

• Gemini 3.5 Flash gains computer use

• Sakana's Fugu-Ultra hits OpenRouter

• Notion adds Claude and Cursor agents

• Exa Connect links agents to data

• Engram raises $98M for AI memory

• Lilian Weng revisits scaling laws

• Plans don't persist in agents

• Tmax opens terminal-agent training

And all the top AI dev news, papers, and tools.

Top Stories

OpenAI Previews GPT-5.6

OpenAI introduced a limited preview of GPT-5.6, a new model family led by Sol, its next-generation frontier model, alongside Terra and Luna for cheaper, higher-volume work.

• Three tiers: Sol is the flagship for ambitious agentic work, Terra delivers GPT-5.5-competitive performance at 2x lower cost, and Luna is the fastest, most affordable option for high-volume tasks.

• Agentic SOTA: Sol sets a new state of the art on Terminal-Bench 2.1, which tests complex command-line workflows requiring planning, iteration, and tool coordination.

• Security frontier: Billed as OpenAI's most capable model for cybersecurity, Sol shifts the performance-efficiency frontier on long-horizon tasks like vulnerability research and exploitation.

• Gated rollout: At the request of the US government, OpenAI is starting with a limited preview for trusted partners in Codex and the API, with general availability planned in the coming weeks.

Blog

Read more