AI Agents周刊：Gemini 4 Argon与OpenAI Dots发布

In today’s issue:

• Google unveils Gemini 4 Argon
• OpenAI launches always-on dots
• OpenAI ships GPT-6.1 Sol
• Anthropic launches Claude Sonnet 5.5
• Earendil ships Pi 1.0 and Pi Durable
• Claude Code adds mods
• DeepSeek open-sources its agent harness
• Manus 2.0 runs on Cascade
• Schema hits 99.2% on ARC-AGI-3
• OpenAI refreshes the Codex CLI

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

Top Stories

Google DeepMind introduced Gemini 4 Argon, its new frontier model for long-horizon coding, enterprise knowledge work, and cyber defense. It is rolling out first to trusted cyber defenders through the Fairwind Program, with paid API customers and Google AI Ultra subscribers next.

• Agentic coding: Argon sets a new state of the art on DeepSWE v1.1 at 77.9%. Inside Google, Argon agents are migrating C/C++ codebases to Rust, up to over 800K lines for the Fuchsia Zircon kernel, and rewrote 32K lines of SIMD code in libgav1 into a memory-safe decoder 2.7x faster than the earlier Rust port.
• Knowledge work: It leads the Vals Index and ranks first on Zapier’s AutomationBench at 51.3%, and scores 91.7% on LVBench for long video understanding.
• 1M output tokens: The output limit rises from 64K to 1M tokens, so the model can work through hundreds of thousands of tokens in a single trajectory.
• Cyber defense: Argon can find, validate, and patch vulnerabilities on its own and ties for first on CWE-bench v1 at 68%. Trusted defenders get it without cyber guardrails, and Google monitors its chain-of-thought and actions and stops execution when needed.
• Pricing: The introductory price is $2 input and $10 output per million tokens with 95% off cached input, rising to $4 and $20 afterward.

Blog

OpenAI introduced dots, always-on agents powered by GPT-6 Astra that run on their own cloud computer, learn from feedback, and keep working toward your goals in the background. They are rolling out in ChatGPT to Pro and Business Premium users in eligible markets, and Enterprise admins can enable a beta.

• Own computer: Each dot has its own cloud computer and browser, connects to over 4,000 apps through plugins, and can run several projects at once. You can open its computer to inspect the work or give it permission to use your laptop.
• Where you work: You can message or call your dot in ChatGPT on desktop, web, and mobile, or reach it in Slack and Teams, and it carries context across every channel. It can also start and manage tasks in Codex and ChatGPT Work.
• Developer workflow: In OpenAI’s example, a dot watches customer feedback, scopes small fixes, builds and tests them, and brings back complete PRs with videos of the changes.
• Controls: Built-in rules decide when a dot acts on its own and when it asks first, and Custom Rules let you allow, require approval for, or block specific actions. Background research runs with read-only tools, and a monitoring system can pause or stop a dot’s work.
• Specialist dots: OpenAI is previewing dots with their own identity and credentials for well-defined roles inside organizations, starting with enterprise pilots and a planned integration with Microsoft Agent 365.

Blog | Docs | Safety

Read more