China May Let Alibaba Buy Nvidia H200s — But Only for Training
Cognition's new SWE-1.7 coding AI model, based on Kimi K2.7 Code and running on Cerebras chips, outputs 1,000 tokens per second and can handle tasks lasting up to 6 hours.
Cognition's new SWE-1.7 coding AI model, based on Kimi K2.7 Code and running on Cerebras chips, outputs 1,000 tokens per second and can handle tasks lasting up to 6 hours.
Mistral AI open-sources Leanstral 1.5, a model for Lean 4 theorem proving that costs about $4 per problem on PutnamBench, outperforming expensive systems. It also found real bugs in Rust code.
Browser Use CLI 3.0 replaces step-by-step AI browser automation with direct Python script execution, cutting token consumption and latency dramatically.
Browser Use CLI 3.0 replaces step-by-step AI browser automation with direct Python script execution, cutting token consumption and latency dramatically.
Alibaba Cloud's Qoder Enterprise Edition can parse codebases with up to 100,000 files, offering team collaboration, security layers, and flexible pricing.
Google’s TabFM can predict customer churn or fraud from structured data without any model training or feature engineering — just feed it a table.
SoftBank has made its first $10 billion payment as part of a $30 billion follow-up investment in OpenAI, bringing its total stake to over $64 billion.
AI coding tools are blurring the lines between programmers, designers, and product managers. Claude Code creator Boris Cherny predicts five new roles will replace the old ones.
Google Antigravity 2.2.1 brings OAuth keychain integration, deadlock fixes, and media playback improvements — but upgrade glitches and the removal of Ctrl+I are frustrating some users.
Anthropic released a policy paper calling China's model distillation a "systematic industrial espionage" and urging Congress to criminalize the practice, warning the US advantage is shrinking to just months.