OpenAI Introduces GPT-6 Sol and GPT-6 Luna with 50% Lower API Prices

OpenAI has announced GPT-6 Sol and GPT-6 Luna, two models positioned as faster and more affordable extensions of the GPT-6 family. According to OpenAI’s announcement, both models use training methods similar to GPT-6 Astra, which OpenAI describes as its strongest overall and most aligned model.

The clearest change is pricing. Per 1 million tokens, GPT-6 Sol input falls from $4 to $2 and output from $20 to $10. GPT-6 Luna input falls from $0.20 to $0.10, while output falls from $1.20 to $0.50. OpenAI characterizes each reduction as 50% versus the corresponding GPT-5.6 promotional pricing.

For visual creators building image or video workflows, the practical implication is room for more iterations, agent steps, and context-heavy production pipelines within the same budget. That is an AiPix analysis rather than a claim that these models create better visual assets directly; the announcement focuses on general intelligence, coding, computer use, and agent workflows.

OpenAI reports gains across several evaluations. In AutomationBench 1.0.6, GPT-6 Sol scored 33.2% at xhigh effort with a $0.27 cost per task, compared with 30.3% for GPT-6 Astra at low effort, 26.9% for Claude Opus 5 at max effort, and 31.4% for Claude Fable 5.1 with Opus 5 fallback at max effort. OpenAI notes that the Claude Fable 5.1 figure excludes fallback costs, which appeared in about 40% of tasks.

On Agents’ Last Exam, GPT-6 Sol scored 56.4% at max effort and cost 60% less per task than Claude Opus 5’s highest result. On DeepSWE v1.1, Sol scored 68.8%, compared with 69.9% for Claude Fable 5 at xhigh effort, at about 80% lower task cost. GPT-6 Luna scored 66.6% at max effort; OpenAI says its cost was 93% lower than Opus 5 and 96% lower than Fable 5 in those comparisons.

OpenAI also reports that Sol makes about half as many errors as its predecessor in an internal factuality evaluation based on de-identified, user-flagged real-world conversations. The company cautions that these conversations do not represent typical usage. In OSWorld 2.0, Sol reached 60.5% at xhigh effort versus 60.3% for Claude Opus 5 at medium effort, at about 80% lower task cost. Luna at max effort reportedly surpassed GPT-5.6 Sol at medium effort for one-tenth the cost.

Prompt caching improvements are another notable change. OpenAI says cached input tokens receive a 90% discount, while controls for reasoning effort and tool availability preserve earlier context for reuse. 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 Luna in the desktop app. They are not yet available in Chat and are offered through the API as gpt-6-sol and gpt-6-luna. OpenAI says evaluations may differ from production ChatGPT because of system prompts and available tools, and competitor results come from public reports.