OpenAI prices GPT-6 Astra at $10/1M input tokens and $50/1M output tokens, matching Anthropic's pricing for Claude Fable 5.1
The Deep View:
Context & Ripple Effects
OpenAI’s API pricing has moved sharply across the GPT-5 line: GPT-5.5 doubled the prior model’s token rates, while GPT-5.6 Sol received a temporary cut to $4 input and $20 output per million tokens. Astra’s $10/$50 rate restores a clear premium tier above that discounted offering.
The price match with Claude Fable 5.1 makes the launch less about undercutting Anthropic than setting a shared top-end reference price. Astra is initially available through OpenAI’s Daybreak access program, limiting the immediate buyer set.
First-order effects
- OpenAI’s Daybreak customers pay $10 per million input tokens and $50 per million output tokens for Astra, versus the temporarily discounted $4/$20 rates for GPT-5.6 Sol; workload routing becomes an immediate cost decision.
- Anthropic loses list-price differentiation against Astra at the stated Claude Fable 5.1 rates, leaving buyers to distinguish the two offerings on factors beyond posted token prices.
Second-order effects
- Enterprise AI teams evaluating OpenAI and Anthropic face a like-for-like top-tier token price, concentrating procurement tests on output quality, reliability, and the token volume required to complete a task.
- OpenAI’s lower-priced GPT-5.6 variants and premium Astra tier give customers an incentive to reserve the higher rate for work where the performance difference offsets the additional inference spend.
Third-order effects
- If leading providers continue to converge on headline token rates for frontier models, competition shifts toward effective cost per completed task rather than the input and output price card alone.
- A widening spread between discounted general models and premium frontier models would reinforce a segmented API market, with model routing becoming a central control over enterprise AI budgets.
The trend: Frontier-model vendors are pairing premium capability tiers with lower-cost alternatives, making workload-specific inference economics more important than a single token price.