GPT-6.2 Astra API Price Cut: $8 and $40 Per Million Tokens, and What Changes When You Migrate
GPT-6.2 Astra, released by OpenAI on 17 November 2026, costs $8 per million input tokens and $40 per million output tokens on the API, 20% less than GPT-6 Astra at $10 and $50. Cached input falls further, from $1 to $0.40, while GPT-6.2 Sol keeps GPT-6.1 Sol's $2 and $10 rates, so teams running Astra on long, repeated prompts gain the most from switching.
Key facts
- 17 November 2026: OpenAI released
gpt-6.2-astraandgpt-6.2-solon the API and in ChatGPT for Plus, Pro, Enterprise and Edu users; Enterprise and Edu need a workspace admin to turn it on. - GPT-6.2 Astra Standard rates per million tokens: $8 input, $0.40 cached input, $10 cache writes (30-minute retention), $40 output.
- GPT-6 Astra Standard rates per million tokens: $10 input, $1 cached input, $12.50 cache writes, $50 output.
- GPT-6.2 Sol: $2 input, $0.10 cached input, $10 output per million tokens, identical to GPT-6.1 Sol.
- Both new models: 1,050,000-token context window, 922,000 maximum input tokens, 128,000 maximum output tokens.
- Above 272K input tokens, input and cache rates double and output costs 1.5 times the short-context rate, for the whole request.
What the Astra cut does to a real bill
Take a workload of 10 million input tokens and 2 million output tokens a month with no caching. On GPT-6 Astra that is $100 of input plus $100 of output, $200 in total. On GPT-6.2 Astra it is $80 plus $80, $160. The saving is the headline 20%.
Caching changes the picture. If 8 million of those 10 million input tokens are cache hits, GPT-6 Astra bills $20 for fresh input, $8 for cached input and $100 for output, $128. GPT-6.2 Astra bills $16, $3.20 and $80, $99.20, about 22.5% less. The cached rate drops 60%, so agent loops and RAG pipelines that resend a long system prompt or document set gain more than chat traffic does. Cache writes also fall from $12.50 to $10 per million. If the difference between these token classes is new to you, our guide to reading LLM API pricing walks through each line.
Sol did not get cheaper, but it may replace Astra for some work
GPT-6.2 Sol keeps the same price as GPT-6.1 Sol, so a Sol user saves nothing per token by upgrading. The case for Sol is substitution. OpenAI says Sol came within 0.5 points of Astra on the FrontierCode 2.0 coding benchmark at about one fifth of the cost, as reported by SiliconANGLE. That is a vendor claim on one benchmark. If you route coding or agent tasks to Astra today, run a side-by-side on your own tasks before paying four times the input rate and four times the output rate for Astra.
Three ways to pay less than the Standard rate
Batch. OpenAI lists GPT-6.2 Astra Batch processing at 50% of Standard rates, which puts output at $20 per million tokens for jobs that can wait. Our comparison of batch versus real-time calls covers which workloads fit.
Background windows. Also from 17 November, the Responses API has two-hour and eight-hour background completion windows in public beta. Token rates are 30% lower than Standard for the two-hour window and 40% lower for the eight-hour window, and both GPT-6.2 models qualify. This sits between real time and Batch for tasks such as overnight report generation.
Stay under 272K input tokens. A single request over that threshold is billed at double input and 1.5 times output for the full request, which more than cancels the 20% cut. Chunking long documents keeps you on the short-context rate.
The reverse also applies: Fast mode costs 2 times Standard, and Ultrafast costs 3 times Standard through 28 February 2027, then 6 times from 1 March 2027.
What changes in your code when you switch
OpenAI's GPT-6.2 guide lists behaviour changes that can raise cost or lower quality if you only swap the model name:
- The Responses API now drops reasoning items after edits to earlier conversation history, such as changing an earlier developer message, a tool call or the model mid-conversation. The model then repeats work and uses more tokens. Keep the conversation prefix unchanged between turns.
- Top-level
instructionsare now appended after the input. Move reusable instructions into a developer message as the first input item so they stay cacheable, which matters given the new $0.40 cached rate. - Tool calling requires the Responses API; Chat Completions does not support it for these models.
- Remove
temperature,top_pandtop_logprobs. If you set reasoning effort tononeorminimal, uselow; the API rejectsnoneon Astra.
What to do this week
- Pull last month's Astra usage and split it into fresh input, cached input and output. Apply the new rates to see your actual saving rather than the headline 20%.
- Check how many requests cross 272K input tokens. Those are your most expensive calls on either version.
- Move reusable instructions out of the top-level field before you switch, or your cache hit rate may fall.
- Test GPT-6.2 Sol on the tasks you send to Astra, using a small eval on your own data.
- If you do vulnerability research, note that OpenAI says both models launch with tighter cybersecurity safeguards that may restrict some legitimate work.
Cheaper frontier tokens also shift the comparison with other providers. Earlier this month we covered GPT-5.6 Sol promo pricing ending on 30 November, and the full article archive tracks other price moves.
Questions people ask
How much cheaper is GPT-6.2 Astra than GPT-6 Astra?
Input and output are 20% cheaper ($8 and $40 against $10 and $50 per million tokens). Cached input is 60% cheaper ($0.40 against $1), and cache writes fall from $12.50 to $10.
Did GPT-6.2 Sol get cheaper?
No. GPT-6.2 Sol costs $2 input, $0.10 cached input and $10 output per million tokens, the same as GPT-6.1 Sol.
What breaks when I switch to GPT-6.2 in the Responses API?
Edits to earlier conversation history now drop reasoning items, top-level instructions move after the input, tool calling needs the Responses API, and sampling parameters such as temperature and top_p must be removed.
Is there a cheaper way to run GPT-6.2 on jobs that can wait?
Yes. Batch is 50% of Standard for Astra, and the Responses API background windows, in public beta, are 30% lower for two hours and 40% lower for eight hours.
Sources