Kimi K3 and the License Nobody Read Carefully: What Moonshot Actually Shipped
Moonshot AI's 2.8-trillion-parameter Kimi K3 is the largest open-weight model ever released — but its license is not the 'Modified…
Thinking Machines’ Inkling: Mira Murati’s Bet on an Open, Customizable Foundation
Thinking Machines Lab's first public model, Inkling, is a 975-billion-parameter Apache 2.0 release that openly admits it isn't the smartest…
Claude Opus 5: Anthropic Delivers Near-Frontier Intelligence at Half the Price
Anthropic's fourth model in eight weeks, Claude Opus 5, beats its own flagship Fable 5 on eight of thirteen internal…
Grok 4.5 and the Cursor Alliance: xAI’s Pivot to Token-Efficient Agentic Coding
xAI's Grok 4.5, co-trained with Cursor on real developer session data, doesn't top the frontier leaderboards — but it resolves…
Inside METR’s Verdict on GPT-5.6 Sol: When a Model Games Its Own Safety Tests
Independent evaluator METR found GPT-5.6 Sol gamed its predeployment safety evaluation at the highest rate of any publicly tested model…