2026-11-09模型发布Anthropic releases Claude Fable 5.2 and Mythos 5.2: 20% cheaper, with breaking API changesFable 5.2 is available to all API customers at $8/$40 per million tokens, and Mythos 5.2 stays limited to vetted organizations. The benchmark gains come from Anthropic's own testing, and the API changes need checking before you switch.阅读约5分钟→
2026-11-09基础设施早间NVIDIA opens NVLink Fusion to Korean NPUs, with GPUs for prefill and Rebellions chips for decodeRebellions will connect its R200 inference chips to NVIDIA's interconnect and SK Telecom will deploy the racks, but the systems aren't expected to be deployment-ready until 2028.阅读约3分钟→
2026-11-08其他下周前瞻Week ahead: CoreWeave, Nebius, Applied Materials and Cisco test AI infrastructure demandCoreWeave reports Tuesday and three more AI infrastructure companies report Thursday. Anthropic's IPO price range is still blank, and two developer deadlines arrive November 30.阅读约3分钟→
2026-11-07每日摘要一周回顾This week, AI labs gated their cyber models and rebuilt the plumbing under coding agentsNovember 1–7 brought no new general-purpose flagship from the big US labs. It brought gated security models, controls inside the agent loop, coding-agent infrastructure and two acquisitions, benchmarks of agent failure modes, and one open-weight model delivered on schedule.阅读约5分钟→
2026-11-06研究Ai2 releases ToyBlocks, and finds agents write better tool descriptions than expertsAn open robot-learning simulator built on blocks and a single arm, plus a result any MCP developer can test: self-written skill descriptions beat expert ones.阅读约4分钟→
2026-11-05安全OpenAI releases GPT-6 Cyber, a gated exploit-hunting model it rates Critical for cyber riskThe GPT-6 Astra variant is API-only for approved Daybreak Red organizations, costs $20/$100 per million tokens, and comes with a system card full of vendor-run evaluations.阅读约4分钟→
2026-11-04商业OpenAI and Meta buy coding-agent tooling on the same dayOpenAI agreed to acquire Pierre, the startup behind Code Storage, to strengthen Codex, and Meta bought Conductor, which runs coding agents in parallel. Neither disclosed a price.阅读约3分钟→
2026-11-04模型发布早间Tencent launches Hy4 with vision and computer use, but the open weights are still to comeTencent's flagship model gains image understanding and agentic computer control at the preview's price. The weights and system cards are promised "in the coming days".阅读约4分钟→
2026-11-03基准测试Ai2's StateBench finds frontier models lose track of changing values in long inputsA new benchmark separates finding a fact from following it through a chain of updates, and frontier models fail the second as documents grow.阅读约4分钟→
2026-11-03开发者工具早间Hugging Face proposes CPU-sized contracts to catch silent Transformers regressionsModel Integration Contracts pair shrunk model fixtures with human-reviewed baselines. In a replay, one flagged a UMT5 training bug that sat in main for almost two months.阅读约3分钟→
2026-11-02开发者工具Anthropic adds message threads to the Claude API so long agent loops stop resending everythingA new public beta lets Messages API clients send only the newest turn. Anthropic says median time to first token falls about 35% on 500k–1M-token requests. Token usage and pricing stay the same.阅读约4分钟→
2026-11-01其他下周前瞻Week ahead: AI chip and software earnings, a Warren deadline for Anthropic, and Kimi K3.1 Pro weightsPalantir, AMD, Supermicro and Qualcomm report between Monday and Wednesday, and Moonshot says Kimi K3.1 Pro's open weights land on November 4.阅读约3分钟→