Same introductory price as Argon, $2 and $10 per million tokens until 30 November 2026, then $4 and $20
Google says Argon stays the better fit for most coding and knowledge work, and that Carbon uses more tokens to reason deeply, so cost per task is not cost per token
Google reports first place on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, and 59 on the Artificial Analysis Intelligence Index v5.0 at announcement
The cyber-permissive variant is limited to the Fairwind Program, and the public model passed 92% of Google's 375 task agentic honesty test against 70% for Argon
Best for the hardest multi-step coding and research work.
Output up to 30% faster than Fable 5.1 at $8 and $40 per million tokens, against $10 and $50, with an optional fast mode up to 2.5x quicker at $16 and $80
49.1% on FrontierCode 2.0 at top effort in Anthropic's table, against 42.8% for GPT-6 Astra
Same model as Mythos 5.2, which is limited to vetted cyber and life sciences organisations. Fable carries the cyber and biology safeguards
Anthropic says 1,000 agents formalised the proof of Fermat's Last Theorem in Lean in 30 hours, a result from its own prerelease testing