OpenAI发布GPT-5.5，专为Agent任务设计

In today’s issue:

• OpenAI ships GPT-5.5
• DeepSeek open-sources V4 Preview
• Kimi releases K2.6 Agent Swarm
• ACL paper flags diversity collapse in multi-agent LLMs
• Sakana launches Fugu multi-agent beta
• ChatGPT gets Workspace Agents
• Codex adds Chronicle screen memory
• Qwen3.6-27B drops flagship coding dense
• Gemini Deep Research Max lands
• Google unveils eighth-generation TPUs

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

Top Stories

GPT-5.5

OpenAI released GPT-5.5, a new class model built specifically for agentic work. It is designed to understand complex multi-step goals, use tools, check its own work, and carry tasks through to completion, and is now powering both ChatGPT and Codex.

• Agentic-first design: GPT-5.5 targets messy, multi-part jobs and is tuned to plan, invoke tools, navigate ambiguity, and keep going until the task is done rather than stopping at a single response.
• Strongest gains where it matters: The biggest jumps are in agentic coding, computer use, knowledge work, and early scientific research, with ChatGPT using full-stack inference improvements to serve the model faster per token.
• GPT-5.5 Pro for hard jobs: A new GPT-5.5 Pro tier is rolling out to Pro, Business, and Enterprise users for demanding tasks, with efficiency gains that make Pro a practical default on long reasoning runs.
• Rollout: Available today in ChatGPT and Codex for Plus, Pro, Business, and Enterprise users, with Pro limited to paid and enterprise tiers.

Blog

Read more