GPT-6 Astra, GPT-6.1 Sol, Gemini 4 Argon, and Claude Fable 5.1 all shipped within a month of each other, but their prices per task differ far more than their benchmark scores do.
Astra and Fable 5.1 share the same list price, $10 per million input tokens and $50 per million output tokens, while Sol and Argon cost one-fifth that. The gap that matters most for AI agents is cached-input pricing, the discounted rate for resending the same system prompt or tool instructions repeatedly: Astra charges $1.00 per million cached tokens, four times Fable 5.1's rate and ten times Sol's.
Argon remains the outlier structurally, with a 1 million-token output limit versus 128,000 for the other three, though it's only available through Google's limited Fairwind testing program for now, and its intro pricing is set to double later.
Key Capabilities:
- Price split: Astra and Fable 5.1 cost 5x more per token than Sol and Argon.
- Cache matters for agents: Astra's cached-input rate is 10x pricier than Sol's.
- Access gap: Argon is Fairwind-program-only; the other three are broadly available via API.