OpenAI · Model release
GPT-6 Sol and Luna: GPT-6 training at half the GPT-5.6 price
OpenAI's second GPT-6 release brings Astra's training advances to a mid-priced tier (Sol) and a volume tier (Luna), each at half the price of the GPT-5.6 model it replaces.
What changed
- GPT-6 Sol costs $2/$10 per MTok (cached $0.20) and GPT-6 Luna $0.10/$0.50 (cached $0.01) — 50% of GPT-5.6 Sol's $4/$20 promo and GPT-5.6 Luna's $0.20/$1.20. Both keep a 1.05M context and 128K max output.
- Input above 272K tokens bills the whole request at $4/$15 (Sol) or $0.20/$0.75 (Luna); Batch and Flex are half price and Fast mode is 2x. The API, ChatGPT Work and Codex got both models the same day.
- OpenAI reports DeepSWE 1.1 at 68.8% for Sol and 66.6% for Luna (max effort, vendor-run). On the independent Artificial Analysis Intelligence Index v4.3.2, Sol scores 47.5 and Luna 37.3 at max effort.
Best fit
- Cost-sensitive reasoning
- High-volume classification and chat
- Codex and OpenAI-native stacks
What to be careful about
- GPT-5.6 Sol and Luna stay live (Sol's $4/$20 promo runs at least through 2026-11-21) — pin the full model id so routing does not move under you.
- GPT-5.6 Terra ($2/$12) has no GPT-6 successor, but it now costs more than GPT-6 Sol; re-run any Terra routing decision.
- The DeepSWE figures are OpenAI's own until Datacurve's independent board lists the models.
Dataset snapshot: GPT-6 Sol is also represented in the checked comparison dataset. Prices and benchmark figures carry their own field dates.