GPT-6 launch: what Astra, Sol and Luna cost
OpenAI released the GPT-6 family in steps during September 2026:
- 3 September: GPT-6 Astra, the top model, broadly available from 4 September.
- 22 September: GPT-6 Sol and GPT-6 Luna.
- 29 September: GPT-6.1 Sol replaced GPT-6 Sol at the same input and output price, with cached input halved from $0.20 to $0.10 per 1M tokens. OpenAI now points GPT-6 Sol users to it.
- 26 October: GPT-6.1 Terra at half GPT-6.1 Sol's price, which OpenAI describes as fast and efficient, for visual reasoning, computer use and natural writing.
All of them have a 1.05M-token context window, cached input at a tenth of the input price or less, and 50% off through the Batch API. Prompts over 272,000 tokens are billed at twice the input price and one and a half times the output price for the whole request.
GPT-6 against GPT-5.6
| GPT-5.6 model | Price now | Replaced by | Price now |
|---|---|---|---|
| GPT-5.6 Sol | $4.00 / $20.00 | GPT-6 Astra | $10.00 / $50.00 |
| GPT-5.6 Terra | $2.00 / $12.00 | GPT-6.1 Sol | $2.00 / $10.00 |
| GPT-5.6 Luna | $0.20 / $1.20 | GPT-6 Luna | $0.10 / $0.50 |
Per 1M input / output tokens. At launch, the middle and small models got cheaper: GPT-6 Sol matched GPT-5.6 Terra on input and cost 17% less on output, and GPT-6 Luna cost half as much as GPT-5.6 Luna on input and 58% less on output. The top model went the other way: GPT-6 Astra cost two and a half times GPT-5.6 Sol's promotional $4 / $20, which OpenAI guarantees until at least 21 November 2026. The GPT-5.6 models are still available through the API.
GPT-6.1 Sol: what changed from GPT-6 Sol
The input and output prices stayed at $2 / $10 per 1M tokens; cached input halved to $0.10. OpenAI describes GPT-6.1 Sol as offering "near-Astra performance" at a lower cost but hasn't published benchmark scores. On Epoch AI’s independent Capabilities Index it scores 168, up from GPT-6 Sol’s 164 and within the margin of error of GPT-6 Astra (169) and Claude Sonnet 5.5 (166), at a fifth of Astra’s price per token.
Before switching, check two changes: GPT-6.1 Sol doesn't support the none or minimal reasoning settings, and tool calling works only through the Responses API, not Chat Completions. Latency-sensitive apps should measure response times first.
What GPT-6 costs on real workloads
For agents, the cached-input rate matters more than the headline price. A coding agent reading 750M input tokens a month, 85% of them from cache, and writing 8M output tokens costs $369 a month on GPT-6.1 Sol, $433 on GPT-6 Sol and $2,163 on GPT-6 Astra. Without caching, the same work on GPT-6.1 Sol would cost $1,580. The agent cost calculator works this out for your own steps and cache rate.
Some organisations get free daily tokens on these models in exchange for sharing their API traffic with OpenAI; the conditions are on our OpenAI API pricing page.
Which GPT-6 model to use
- GPT-6 Luna for high-volume, simple work: classification, extraction, routing and short answers. It is one of the cheapest models from a major lab.
- GPT-6.1 Sol for most production work, including chat, agents and coding. For prompts under 272,000 tokens it costs the same as Claude Sonnet 5.5 per input, cached and output token, so the choice between them comes down to how each does on your tasks.
- GPT-6.1 Terra at half Sol's price, for vision, computer use and writing, or as a cheaper model to test before Sol.
- GPT-6 Astra for the hardest tasks, where a better first attempt saves more than it costs. It is five times the price of GPT-6.1 Sol. On the Arena text leaderboard on 25 September it scored 1478, below Claude Opus 5.5 (1509, provisional at the time) and Gemini 3.8 Flash (1492), so test it on your own tasks before paying for it.
Head-to-head: Claude Opus 5.5 vs GPT-6 Astra · Claude Sonnet 5.5 vs GPT-6.1 Sol · GPT-6 Luna vs Claude Haiku 5.5. Billing, the free tier and ChatGPT plans are covered on the OpenAI API pricing page.
Sources: OpenAI API pricing, GPT-6.1 Sol model page, OpenAI changelog, Epoch AI Capabilities Index (CC BY 4.0).
