- Start
- Sep 22, 202692% CONFIDENCEfrom developers.openai.com
GPT-6 Sol and GPT-6 Luna Released
- OpenAI released GPT-6 Sol and GPT-6 Luna on Tuesday 22 September 2026, extending the GPT-6 generation that began with GPT-6 Astra on 3 September; both went into the API as
gpt-6-solandgpt-6-lunaand started rolling out in ChatGPT Work and Codex for paid plans, with Free and Go users able to try Luna in the desktop app[1][4][5][6] - API prices are half those of GPT-5.6: Sol costs $2 per million input tokens and $10 per million output tokens against $4 and $20, and Luna $0.10 and $0.50 against $0.20 and $1.20; OpenAI told VentureBeat the rates are permanent, not promotional, and credits better caching and inference[4][5][6]
- The launch came about 90 minutes after Anthropic released Claude Opus 5.5, and OpenAI claims both models beat Anthropic's top models on many tasks[5]
- On OpenAI's internal factuality test Sol makes "about half as many mistakes" as GPT-5.6 Sol, and in a coding-deception test its deception rate falls to 1.3% from 10.4% (Luna 2.8% from 9.5%)[5][6]
- On 25 September OpenAI fixed an image-encoding bug that had degraded both models' image understanding in the API and Codex, including computer use[4]
Notable features
- Sol is built for complex coding and agentic workflows; Luna is OpenAI's "most efficient model for focused, high-volume tasks" such as summarizing documents, extracting information and answering quick questions[2][3][5]
- Specifications for both: 1,050,000-token context window (922,000 maximum input tokens), 128,000 max output tokens, text and image input with text output, reasoning effort from none to max; knowledge cutoff April 20, 2026 for Sol and May 18, 2026 for Luna[2][3][4]
- Cached input costs $0.20 per million tokens on Sol and $0.01 on Luna; prompts over 272K tokens are billed at 2x input and 1.5x output[2][3][4]
- OpenAI's reported scores: AutomationBench Sol 33.2% at $0.27 per task against Claude Opus 5's 26.9%; DeepSWE 1.1 Sol 68.8% and Luna 66.6% against Claude Fable 5's 69.9%; OSWorld 2.0 Sol 60.5% against Opus 5's 60.3%[6]
References 690% CONFIDENCEOverall confidence: 90%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
- [1]90%openai.com/index/introducing-gpt-6-sol-and-lunaopenai.com· Posted Sep 28, 2026· Starts Sep 22, 2026· 17% of score
OpenAI's[2][3][4] API changelog entry 'Sep 22' reads 'Released GPT-6 Sol (gpt-6-sol) and GPT-6 Luna (gpt-6-luna)'; TechCrunch[5] (2026-09-22T18:00Z) says the models 'are now available in ChatGPT Work and Codex' and OpenAI's X post the same day says they 'roll out today'.
- [2]90%GPT-6 Sol Model | OpenAI APIdevelopers.openai.com· Added Sep 28, 2026· 17% of score
OpenAI's[1][3][4] model page: 'built for complex coding and agentic workflows', 1,050,000 context window, 922,000 maximum input tokens, 128,000 max output tokens, Apr 20, 2026 knowledge cutoff, reasoning effort none to max, $2 input / $0.20 cached / $2.50 cache writes / $10 output per 1M tokens.
- [3]88%GPT-6 Luna Model | OpenAI APIdevelopers.openai.com· Added Sep 28, 2026· 17% of score
OpenAI's[1][2][4] model page: 'our most efficient model for focused, high-volume tasks', 1,050,000 context window, 128,000 max output tokens, May 18, 2026 knowledge cutoff, $0.10 input / $0.01 cached / $0.50 output per 1M tokens.
- [4]92%Changelog | OpenAI APIdevelopers.openai.com· Published Sep 22, 2026· Starts Sep 22, 2026 ✓· 16% of score
OpenAI's[1][2][3] own API changelog, entry 'Sep 22': 'Released GPT-6 Sol (gpt-6-sol) and GPT-6 Luna (gpt-6-luna)', reasoning models with text and image input and text output, priced per 1M tokens at '$2 input, $0.20 cached input, and $10 output' (Sol) and '$0.10 input, $0.01 cached input, and $0.50 output' (Luna); a 'Sep 25' entry fixes an image-encoding bug that had degraded their image understanding.
- [5]90%OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakestechcrunch.com· Published Sep 22, 2026· Starts Sep 22, 2026· 16% of score
TechCrunch's same-day report (2026-09-22T18:00Z): the 6-series models cost half as much as GPT-5.6 Sol and Luna thanks to caching and inference gains, Sol makes 'about half as many mistakes as its predecessor', Anthropic's Opus 5.5 came out 90 minutes earlier, and the models 'are now available in ChatGPT Work and Codex for most paid accounts'.
Suggest a correction
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.