Caching and the long-context rule in numbers
Astra's cache pricing has two parts: writing a prompt prefix into the cache costs $12.50 per million tokens, and reading it back costs $1. A prefix that is reused even once is already cheaper than sending it twice at the normal $10 rate, so caching pays off quickly for agents and chat apps that resend a long system prompt.
The 272,000-token rule works the other way. A request with 280,000 input tokens is billed at $20 per million for all of its input, not only the part above the line, and its output at 1.5 times the normal rate. Keeping individual requests under the threshold, for example by summarising history or retrieving only the relevant documents, can halve the input cost of very long prompts.
Where to use it, and where not
Astra is available through OpenAI's API and also through Microsoft Azure and AWS Bedrock. Since 22 September it sits above two much cheaper GPT-6 models: GPT-6 Sol at $2 / $10 and GPT-6 Luna at $0.10 / $0.50. A sensible pattern is to route most requests to Sol or Luna and send only the hardest tasks, where a cheaper model fails or needs several attempts, to Astra. Measure cost per completed task rather than per token when deciding.
Tip: Update, 29 September 2026: OpenAI has since released GPT-6 Sol ($2 / $10) and GPT-6 Luna ($0.10 / $0.50), which replace GPT-5.6 Sol and Luna at about half the price. The comparisons below now use them. See our GPT-6 Sol and Luna pricing post for the full breakdown.
OpenAI released GPT-6 Astra on September 3, 2026. It now sits above the GPT-5.6 family as OpenAI's flagship model, and its standard API price is $10 per million input tokens and $50 per million output tokens.
The rate card
- Standard: $10 input and $50 output per million tokens. Cached input costs $1 per million.
- Batch and Flex processing: half price, $5 and $25.
- Fast mode: double, $20 and $100.
- Prompts over 272,000 input tokens reprice the whole request at 2 times the input rate and 1.5 times the output rate.
What a typical request costs
For a request with 5,000 input tokens and 1,000 output tokens, run 10,000 times a month:
Input
5,000 × $10.00 ÷ 1Mequals$0.05000
Output
1,000 × $50.00 ÷ 1Mequals$0.05000
Per month
$0.10000 × 10,000equals$1,000.00
The same workload costs $200 a month on GPT-6 Sol or Claude Sonnet 5 (both $2 / $10), and $10 on GPT-6 Luna at $0.10 / $0.50. Astra is aimed at long, multi-step agent work, where OpenAI says finishing tasks in fewer steps can offset the higher rate. For chat, summaries and classification, the cheaper tiers usually do the job for a fraction of the price.
Tip: Run your own workload through the AI API Cost Calculator and the AI Feature Cost Calculator. Both now include GPT-6 Astra.
Questions people ask
- How much does GPT-6 Astra cost per million tokens?
- $10 for input and $50 for output, with cached input at $1. Batch and Flex processing are half price.
- What happens with very long GPT-6 Astra prompts?
- Requests over 272,000 input tokens are repriced at 2 times the input rate and 1.5 times the output rate for the whole request.
- Is GPT-6 Astra worth it for chat?
- Usually not on price alone. A 10,000-request workload costs $1,000 a month on Astra against $200 on GPT-6 Sol or Claude Sonnet 5 and $10 on GPT-6 Luna.
- How much do GPT-6 Astra cache writes cost?
- $12.50 per million tokens to write to the cache and $1 per million to read it back, against $10 for normal input.
Sources
Get fee-change alerts
One short email when a price our calculators use changes. No other mail, unsubscribe anytime.