AI
Everything we publish on AI: the models we run on our own machines, what the tools send home, the benchmarks we measure ourselves and the claims we check against the code. Start with the investigations, then the guides.
NewsAnthropic discloses a fourth case of a Claude model reaching real systems in a test
Anthropic's alignment assessment discloses a fourth incident of a Claude model reaching real systems during testing, found by scanning 481…
NewsMeta releases Muse Spark 1.3 to developers and lists an open weights release
Meta released Muse Spark 1.3 on September 2 in Muse Code and its API, reporting 20% fewer tool calls than…
NewsSanders and Casar announce the Ban Artificial Superintelligence Act
Sanders and Casar announced the Ban Artificial Superintelligence Act on September 3: a permanent ban, a pause on advanced AI,…
NewsPi coding agent adds GPT-6 Astra two days after OpenAI’s launch
The Pi coding agent added GPT-6 Astra in version 0.85.1 on September 5, two days after launch, and fixed an…
NewsNous Research makes Bot Mode the default in the Hermes Agent desktop app
Hermes Agent 0.21.0 makes Bot Mode the default in its desktop app: named bots, rooms of two to six bots,…
NewsOpenClaw lets every agent read other agents’ sessions by default
OpenClaw 2026.9.2 lets every agent on a shared Gateway read other agents' sessions by default. The change was reviewed as…
NewsAlibaba publishes Qwen3.8-Max weights, keeping vision input for its hosted version
Alibaba published Qwen3.8-Max weights on August 9. The open checkpoint takes text only at 262,144 tokens natively; the hosted version…
NewsMoonshot AI publishes Kimi K3 weights: 2.78 trillion parameters in 1,561 GB
Kimi K3 weights went public on July 27. Hugging Face counts 2.78 trillion parameters in 1,561 GB of files, and…
NewsAnthropic releases Claude Fable 5.1 at the same prices with cheaper cache reads
Anthropic released Claude Fable 5.1 on September 1 at the same $10 and $50 prices, with cache reads cut to…
NewsOpenAI releases GPT-6 Astra with a 99.9% claim that depends on its own harness
OpenAI released GPT-6 Astra on September 3 and claims 99.9% on ARC-AGI-3. ARC Prize measured that only on OpenAI's own…