GPT-6 Sol and Luna at half price
On 22 September 2026 OpenAI released GPT-6 Sol and GPT-6 Luna, two models below GPT-6 Astra (3 September) in capability and price, in ChatGPT Work, Codex and the API as gpt-6-sol and gpt-6-luna. API prices are half the promotional prices of the GPT-5.6 models: Sol from $4 and $20 to $2 and $10 per million input and output tokens, Luna from $0.20 and $1.20 to $0.10 and $0.50.
Why it matters
The previous generation's pattern repeats within three weeks: a flagship first, then cheaper tiers trained with similar methods. The point OpenAI's post leads with is the price cut, and it is made by halving, not by a new tier.
Availability. From 22 September Sol and Luna are available in ChatGPT Work and Codex to Plus, Pro, Business, Enterprise and Edu users; Luna is also in the desktop app for Free and Go users. They are not yet in Chat mode. In the API they are gpt-6-sol and gpt-6-luna; standard prices per million tokens for prompts up to 272K input tokens: Sol $2 input, $0.20 cached, $10 output; Luna $0.10, $0.01 and $0.50. The discount on cache reads is 90%. The rollout in ChatGPT ran through the day. OpenAI's claims about quality. AutomationBench: Sol at max effort beats Claude Opus 5 at max effort at 9% of its cost per task. Agents' Last Exam: Sol 56.4%. DeepSWE v1.1: Sol 68.8% (1.1 points below the best result listed for Claude Fable 5, 69.9%, at about 20% of its cost per task), Luna 66.6%. OSWorld 2.0 offline (partial score): 60.5% against 60.3% for Opus 5 at medium effort. On an internal factuality evaluation Sol makes about half as many errors as its predecessor. Comparisons with competitors use their public reports; where Fable 5.1 scores were unavailable, Fable 5 is used. What the record does not claim. All evaluations are OpenAI's, run in its research environment or through the API; they may differ from what runs in ChatGPT. No independent measurements were read. The post prints no date in its text: 22 September is from the API changelog.