AI 周报把 Gemini 4 Argon 列入本周重大更新
一份 AI 周报汇总 Claude Haiku 5.5、多智能体编排、GPT-6.1 Sol、Meta Muse 等更新,Gemini 4 Argon 也在列但正文被截断。

AI Weekly Report The biggest AI updates this week, from cheaper models to 1,000-agent workflows. Claude Haiku 5.5 • Anthropic’s fastest model, generating up to 200 tokens/sec, with adjustable reasoning effort for the first time in Haiku. • Input starts at $0.10 per 1M tokens for prompts up to 100K tokens. • Max and Team subscribers now get monthly API credits: $100 for Max 5x, $200 for Max 20x and up to $500 for Team. Claude Multi-Agent Orchestration • A lead agent can coordinate up to 1,000 subagents in a single run. • In Anthropic’s test, the workflow found 66 of 70 planted bugs, compared with 14-27 found by a single agent. • Public beta is API-only, not included in subscription usage limits. GPT-6.1 Sol in Codex and Work • Default context window increased from 272K to 500K tokens, around 80% more. • Less context compaction means more details can be preserved in long tasks. • Manual configuration up to 1M tokens was already possible, with no extra long-context charges within the subscription. Meta Muse Spark Open • Open-weight model with 816B parameters, 42B active parameters and a 1M-token context window. • Built for coding, multimodal tasks and complex agent workflows. • Can run across two Mac Studio systems with 1TB combined memory, and supports NVIDIA, AMD and Intel hardware. GPT-Live-1.1 • Voice updates are rolling out across ChatGPT, Codex and dots. • Adds voices for French, German, Hindi, Mexican Spanish, Korean and Japanese. • Improved accents and fewer unnecessary “checking…” interruptions. Kling 4.0 • Generates videos up to 30 seconds, double the previous 15-second limit. • Supports 10 keyframes and up to 15 reference files per prompt. • Adds prompt-based video editing, stereo audio, improved lip-sync and 10-bit HDR in 1080p and 4K. Gemini 4 Argon • Now available to Pro subscribers in AI Studio. • Access is included across Pro subscriptions, not just plans costing the full $20/month. DeepSeek-V4.1-Flash-1017 • Faster, more concise responses help reduce token usage and API costs. • Public Search and Fetch APIs bring web search and page retrieval to developers for $0.75 per 1,000 requests each. • Server-side search and fetch are free for supported coding-agent API workflows until October 31, 2026, 16:00 UTC. The trend is clear: more context, more capable agents and lower costs. The real advantage comes from using them efficiently.