OpenAI launched GPT-6 Astra on September 3, 2026 at a token rate well above the model it replaces. The OpenAI model pricing question for platform leads is not that rate alone: Astra uses far fewer tokens per task, and whether that offsets the increase depends on the workload.
What changed
OpenAI introduced GPT-6 Astra as its “most intelligent and aligned model,” available in the API as gpt-6-astra and through Microsoft Azure and AWS Bedrock.
OpenAI publishes the rates directly: Standard pricing is $10 per million input tokens and $50 per million output tokens, with separate rates for cache reads and writes, and a Fast mode that delivers up to 2x the speed of Standard processing at 2x the Standard price. Biggo Finance sets those rates against GPT-5.6 Sol at $4 and $20 per million tokens, calling it a 2.5x increase, and notes Astra’s rates match Anthropic’s Claude Fable 5.1.
Why B2B teams should care
OpenAI’s case rests on efficiency. On Terminal-Bench Science 0.1 it reports Astra at 64.6% against 52.6% for Claude Fable 5.1, at approximately 31% lower estimated API cost, and 61.1% at a lower-cost setting against Sol’s best result of 22.4%.
Independent measurement splits by workload. Artificial Analysis, as reported by Biggo Finance, found Astra used roughly one-third the tokens of Sol at maximum reasoning and put it on the Pareto frontier for coding agents, where task cost ran nearly identical to Sol while scoring two points higher on the Coding Agent Index. On the Intelligence Index that efficiency did not pay for itself: both models scored 61, and Astra’s per-task cost ran 75% higher than Sol’s for that equal score.
Who is affected
Biggo Finance reports Pro, Enterprise, and Business Premium users gained access in ChatGPT Work and Codex within two days of launch, with API availability following. OpenAI says Astra usage falls within existing subscription allowances, with credits purchasable for more, and that enterprise administrators must enable Astra because access is off by default at launch.
What teams should check now
Teams should compare cost per successful task rather than cost per token, and split the two workload families the source pack divides on: agentic coding and terminal tasks, where the token savings offset the higher rate, and general reasoning, where they did not.
Budget models should also carry what sits outside the headline rate — the separate cache read and write rates, and Fast mode at 2x the Standard price.
What remains unclear
- Not yet confirmed: How the published token rates translate into production spend, since both OpenAI and Artificial Analysis report benchmark-estimated API cost rather than customer billing data.
- Not yet confirmed: OpenAI’s cache read and write rates and its Fast mode rate in absolute terms, which the announcement gives only relative to Standard pricing.
- Not yet confirmed: Whether Astra’s coding-agent economics hold outside coding, terminal, and computer-use tasks.
What to watch next
Headline cost figures from public demonstrations need reading carefully. Tom’s Hardware reported that Astra autonomously completed Valve’s Portal across roughly 24 hours and 3,336 tool calls at a headline API cost of $571.18 — but the enthusiast who ran it, CozyBlaze, has since clarified the cost was covered by their $200 Codex Pro subscription, and says the run is not a benchmark.
