METR estimates faster benchmark progress at Anthropic, but uncertain attribution and redacted evidence limit what the result proves.
Gemini Omni 1.2 Flash adds 15-second video and lower output pricing. Access restrictions, token billing and evaluation limits matter before adoption.
GPT-6 Cyber begins a restricted rollout for approved security teams. Its benchmark gains need to be read alongside access requirements and mixed external results.
TwIL-8B's balanced checkpoint targets shorter reasoning. Its noncommercial license and mixed benchmark methods are essential context for researchers.
Soniox’s new speech models are active, with real-time text context and echo filtering. Benchmark and output-format caveats matter before migration.
IndiaAI’s November 3 model launch pairs Soket’s EKA-1 with ZenteiQ’s compact Dhīr models. Access routes and benchmark evidence need careful reading.
Images 2.6 adds GIFs, API markup and SVG output. The release’s workflow claims and developer-reported safety results call for different kinds of evaluation.
Moonshot’s public model files turn a promised release into a deployment option. The custom license and infrastructure requirements need careful review.
Sarvam Prism scores predefined answers instead of generating text. Compare its developer-reported benchmarks, API access and local hardware requirements.
OpenAI’s proposed Pierre deal highlights code review and Git workflows. Existing tools and future integration are different milestones.
Hy4’s access options and a cautious reading of Tencent’s finance-task comparison.
New chemistry tools connect AI workflows with structured analysis, while proposed compounds still need experimental checks.
New engineering collaborations put verification and manufacturability at the center of AI-assisted design.
New university grants support practical AI research, following Google's Singapore research-hub announcement.
What OpenAI’s planned shared allowance means—and why Instant chat is an important exception.
Receive the latest AI news
