358B / ~32B (MoE)
MIT
Latest flagship model with major improvements in coding and creative writing; Core Coding: 73.8% SWE-bench (+5.8%), 66.7% SWE-bench Multilingual (+12.9%), 41% Terminal Bench 2.0 (+10%), 84.9% LiveCodeBench v6; Vibe Coding: cleaner/more modern webpages and better-looking slides with accurate layout/sizing; Complex Reasoning: 42.8% HLE with tools (+12.4%), 95.7% AIME 2025, 97.1% HMMT Feb 2025, 93.5% HMMT Nov 2025; Tool Using: 87.4% τ²-Bench (SOTA), 67.5% BrowseComp with context management; Creative Writing: nuanced/vivid prose with sensory details (scent, sound, light), consistent character archetypes and worldbuilding adherence, natural tension-filled plot progression, better context/sentiment capture, reduced politeness/redundancy, improved multi-turn context memory; Thinking Control: Turn-level Thinking (explicit on/off toggle), Preserved Thinking (retains thinking blocks across turns), Interleaved Thinking (thinks before every response/tool call); integrated with Claude Code, Kilo Code, Roo Code, Cline; available via API and OpenRouter.