OpenAI introduces GPT-6 Sol and Luna

The company said the models were developed using methods similar to those behind GPT-6 Astra, which OpenAI introduced earlier this month. While Astra remains the company's most capable model, Sol and Luna are aimed at users and developers who need advanced performance at greater speed and lower cost.

OpenAI said the new models improve on their GPT-5.6 counterparts in professional work, factuality, coding, computer use and communication, while also reducing API prices by 50%.

Under the new pricing, GPT-6 Sol costs $2 per 1 million input tokens and $10 per 1 million output tokens, compared with $4 and $20 respectively for GPT-5.6 Sol. GPT-6 Luna costs $0.10 per 1 million input tokens and $0.50 per 1 million output tokens, down from $0.20 and $1.20 for GPT-5.6 Luna.

In professional workflow testing, OpenAI said GPT-6 Sol achieved a 33.2% score on AutomationBench at its highest reasoning setting, compared with 26.9% for Claude Opus 5 at maximum effort. The company reported that Sol's cost per task was $0.27, while Claude Opus 5 cost 11.1 times more in the same evaluation.

GPT-6 Sol also scored 56.4% on Agents' Last Exam, which tests AI agents on complex professional workflows across 55 sub-industries. OpenAI said the model achieved this result at 60% lower cost per task than the highest Claude Opus 5 result in the evaluation.

The company also reported improvements in factual accuracy. According to OpenAI's internal evaluation based on de-identified conversations in which users had previously flagged factual errors, GPT-6 Sol makes about half as many mistakes as GPT-5.6 Sol. GPT-6 Luna also showed substantial improvement, with higher-effort settings matching GPT-5.6 Sol at a much lower cost.

For coding, GPT-6 Sol scored 68.8% on the DeepSWE v1.1 evaluation of complex software engineering tasks, compared with 69.9% for Claude Fable 5 at its highest reported setting. OpenAI said Sol achieved the result at approximately 80% lower cost per task. GPT-6 Luna scored 66.6%, with OpenAI reporting significantly lower costs than comparable competitor models.

The new models also improve computer-use performance. On the OSWorld 2.0 offline evaluation, GPT-6 Sol achieved a 60.5% score at its highest reasoning setting, compared with 60.3% for Claude Opus 5 at medium effort, while costing approximately 80% less per task.

OpenAI said Sol and Luna also adopt the communication improvements introduced with Astra, including clearer responses, less jargon, fewer unnecessary details and shorter answers while retaining substantive information.

GPT-6 Sol and GPT-6 Luna are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Free and Go users can access GPT-6 Luna through the desktop app. The models are not yet available in Chat, although OpenAI said it plans to roll them out there gradually.

Earlier, Qazinform News Agency reported that Sam Altman challenged estimates of ChatGPT’s water use.