Anthropic says Claude leads about a quarter of its R&D work
Anthropic reports that Claude leads 26% of its AI R&D task categories under human supervision, a workflow-automation measure rather than autonomous model development.
OpenAI launches Astra for Law with a U.S. legal search index
OpenAI's Astra for Law combines GPT-6 Astra with U.S. legal retrieval and firm controls, reporting 54.0% correctness on a private benchmark but no public citation-error rate.
Discourse patches a libheif upload flaw after AI-assisted exploit claims
Discourse confirmed and patched a high-severity libheif image-upload vulnerability, while the claimed OpenAI account and repository escalation remains only partly public and technically unresolved.
Qwen3.8-Omni-Flash is a long-context media-analysis API
Alibaba's Qwen3.8-Omni-Flash accepts text, images, audio and video with a 1M-token context window, but its documented output is text and it is not a downloadable real-time voice agent.
Bonsai 2 ships a 27B local model in a 5.95 GB download
Prism's ternary Bonsai 2 compresses its smallest 27B language-weight pack to 5.95 GB, but the headline is a disk-size claim rather than a full runtime or VRAM requirement.
Z.ai says GLM helped build its own inference infrastructure
Z.ai says a GLM-5.3-powered agent helped bring a serving stack for more than 100,000 Chinese accelerators to production in under two weeks and roughly tripled throughput under engineer supervision.
Figure reports zero-shot household transfer across 30 homes
Figure says Helix 2.5 completed three predefined household behaviors in unfamiliar homes 56% of the time, demonstrating environment transfer rather than a general autonomous housekeeper.
Laertes leads Metaculus's FutureEval bot tournament
FutureSearch lists the AI forecasting bot laertes first in Metaculus's Summer FutureEval tournament, but the bot's model, tools, cutoff and leakage controls remain undisclosed.