AI News Today
The 6 top stories of the 120 AIBriefs curated today.
DeepSeek-V4.1-Flash packs 552B backbone parameters, native image and text input, and up to a 1M-token context window. The YOCO architecture cuts GPU memory and prefill latency; Bloomberg reports pricing as low as a fraction of a cent per million tokens.
OpenAI's most capable model for business adds advanced reasoning, computer use, and stronger writing and design judgment. Astra began rolling out September 3, eight weeks after GPT-5.6 Sol launched on July 9.
Fable 5.1 scores 55.8% on Terminal-Bench 4.0 in Claude Code, ahead of Fable 5 (42%) and Opus 5 (52.3%). Priced the same as Fable 5, with API cache reads cut 75% to $0.25/MTok.
Amodei's 3,500-word essay proposes a three-part plan to slow AI development; Anthropic unilaterally commits to giving third-party evaluators permanent employee-level access to its systems. Sam Altman agreed and said OpenAI will do the same, with Demis Hassabis, Elon Musk and Satya Nadella also signaling support.
Coxon's resignation post drew 143 million impressions in 24 hours; he told CNN the industry is "gambling with our lives" and Fox News that AI is "possibly the most dangerous technology that humanity has ever created." An Anthropic safety researcher said there is a greater than 10% chance AI could "kill all humans."
OpenAI asked lawmakers in recent weeks whether rival labs could legally coordinate a slowdown without violating antitrust law, per WIRED. Chief scientist Jakub Pachocki's blog post urged "coordinating to slow down future development" and predicted "voluntary slowdowns to become commonplace."
Also today
102 more stories today — open the live feed.
Get tomorrow's AI brief in your inbox