2026-10-22开发者工具OpenAI makes its Decisions API generally available: typed answers from gpt-6-luna, with no output-token chargeThe classification-and-routing endpoint leaves beta with up to 400 questions per request, JSON input, per-question images and prompt-cache discounts. The 10x speed figure is OpenAI's own.阅读约4分钟→
2026-10-21开源Recursive releases Flint, an open-weight coding model built to run agents on a 48 GB laptopRecursive's first model has 39.8 billion parameters with about 2.5 billion active, ships under Apache 2.0, and comes with quantized builds for Apple Silicon and NVIDIA. Its benchmark and speed figures are the company's own.阅读约4分钟→
2026-10-21基础设施早间Google Cloud unveils A5X on NVIDIA Vera Rubin, with general availability set for NovemberGoogle's first Vera Rubin NVL72 instances have one early customer and a fabric Google says can span up to 960,000 Rubin GPUs across sites. Pricing, specs and regions are still unpublished.阅读约4分钟→
2026-10-20开发者工具OpenAI adds a 12-hour prompt cache to the Responses API, at twice the write priceA new ttl option keeps cached prefixes eligible for discounted reads for at least 12 hours instead of 30 minutes. Writes cost 2.5x the input price, so it only pays off when the same prefix is reused across long gaps.阅读约4分钟→
2026-10-20基础设施早间Microsoft sets Ignite GA for Foundry Local and GitHub Enterprise Local in its sovereign cloud pushNew EU oversight of Microsoft engineer access is rolling out now; the on-premises AI pieces are scheduled for Ignite, and key details such as the model catalog are still unpublished.阅读约4分钟→
2026-10-19开发者工具doubleAI launches Maxwell, a model that rewrites programs' machine code while they runThe Shashua-led lab claims 38x faster data pipelines and 4.5x faster inference kernels, with formal proofs that behavior is unchanged. All figures are its own.阅读约3分钟→
2026-10-18其他下周前瞻Week ahead: CFTC's compute-futures comments close, OpenAI retires gpt-4 and o1A Tuesday deadline on whether GPU hours and inference tokens can back futures contracts, IBM's results Wednesday, and model shutdowns at Google and OpenAI on Thursday and Friday.阅读约4分钟→
2026-10-17每日摘要一周回顾Open weights reached frontier scale this week. The scorecards stayed in-house.Meta and Google priced frontier APIs at the same point, Meta, NVIDIA and Reflection released large open weights under very different licenses, and RTX Spark laptops went on sale. Nearly every benchmark was the vendor's own.阅读约5分钟→
2026-10-16开源Reflection releases Beam's weights: 501B parameters, Apache 2.0, and benchmarks it graded itselfThe first model from the Nvidia-backed U.S. lab is now downloadable under a standard permissive license. Its efficiency pitch is promising but still unverified.阅读约4分钟→
2026-10-16基础设施早间NVIDIA RTX Spark laptops go on sale from $2,599, and the entry model has 24GB, not 128GBNine Arm-based Windows laptops from six PC makers are available today. The headline 128GB and 1-petaflop figures describe the top configurations.阅读约3分钟→
2026-10-15模型发布Meta releases Muse Spark Open: 816B open weights, a 1M-token window, and a license with a revenue clauseThe open-weight Muse Spark is built for coding agents and can be self-hosted, but it needs hundreds of gigabytes of memory, and its capability claims are Meta's own.阅读约4分钟→
2026-10-14安全早间NVIDIA pledges $250 million in compute for U.K. agentic AI safety research, with terms still to be agreedThe five-year donation is GPU access worth $50 million a year, routed through U.K. cloud partners. Oxford's BOLD, UCL's SOFAIR and Imperial get a share.阅读约3分钟→