The Pulse
OpenAI Cuts GPT-6 Sol and Luna API Prices; Luna Output Falls 58.3%
OpenAI introduced GPT-6 Sol and GPT-6 Luna with lower API prices, strong results across workflow, coding and computer-use tests, and availability in ChatGPT Work, Codex and the API.
AI.info Team ·
OpenAI launches GPT-6 Sol and GPT-6 Luna on September 22, 2026, expanding the company’s newest model family beyond the flagship GPT-6 Astra. The release places the two models in the OpenAI API, Codex and ChatGPT Work, with lower API prices than the promotional rates for their GPT-5.6 predecessors. Sol’s input and output prices are halved, while Luna’s input price is halved and its output price falls by 58.3%.
GPT-6 Sol now costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20. GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, compared with $0.20 and $1.20 for GPT-5.6 Luna. OpenAI says the prices apply per 1 million tokens.
The launch post describes Sol as the higher-capability option for demanding work and Luna as the lower-cost model for larger volumes of everyday tasks. OpenAI says both models use training methods similar to GPT-6 Astra, while improvements in inference and caching reduce the cost of serving them.
Sol and Luna join GPT-6 Astra
OpenAI introduced GPT-6 Astra earlier in September as its highest-capability model. Sol and Luna extend that family to users and developers who may not need Astra’s full performance or price.
“While the most demanding and important projects still call for Astra’s full depth, work happens at different scales, rhythms, and budgets.”
OpenAI, the publisher of the announcement, positions Sol and Luna as models for work that happens at different levels of complexity, speed and budget rather than as replacements for Astra.
The release also changes the model family’s pricing structure. GPT-6 Sol is priced at one-fifth of Astra’s standard API rates for input and output tokens, while GPT-6 Luna is priced at one-hundredth of Astra’s input rate and one-hundredth of its output rate. Those comparisons use Astra’s published standard rates of $10 per million input tokens and $50 per million output tokens.
OpenAI claims large gains on workflow tests
OpenAI reports that GPT-6 Sol at its xhigh effort setting scores 33.2% on AutomationBench at a cost of $0.27 per task. The company says that result exceeds GPT-6 Astra at low effort, Claude Opus 5 at maximum effort and Claude Fable 5.1 with Opus 5 fallbacks, though OpenAI also notes that the Fable 5.1 figure does not include the cost of those fallbacks.
On Agents’ Last Exam, OpenAI says GPT-6 Sol reaches 56.4% while running at 60% lower cost per task than Claude Opus 5’s highest score in the company’s comparison. Agents’ Last Exam evaluates long-running workflows across 55 professional sub-industries, according to the announcement.
Those figures come from OpenAI’s own evaluation presentation. The company says competitor results were taken from publicly available reports and warns that model evaluations run through its research environment or API can differ from production ChatGPT because of differences in prompts, tools and system settings.
Lower prices target coding and computer-use workloads
OpenAI says GPT-6 Sol scores 68.8% on DeepSWE v1.1 at maximum effort, within 1.1 percentage points of Claude Fable 5’s highest reported result of 69.9% at xhigh effort. OpenAI estimates that Sol costs about 80% less per task in that comparison.
GPT-6 Luna scores 66.6% on the same software-engineering evaluation at maximum effort, according to OpenAI. The company says Luna costs 93% less per task than Claude Opus 5 and 96% less than Claude Fable 5 in its comparison.
For computer use, OpenAI reports that GPT-6 Sol at xhigh effort scores 60.5% on the offline version of OSWorld 2.0, compared with 60.3% for Claude Opus 5 at medium effort. OpenAI estimates that Sol costs about 80% less per task in that test, while GPT-6 Luna at maximum effort exceeds GPT-5.6 Sol at medium effort for one-tenth of the cost.
OpenAI says factuality improves with the price cuts
The company also claims that GPT-6 Sol makes about half as many mistakes as GPT-5.6 Sol on an internal factuality evaluation based on de-identified ChatGPT conversations where users had flagged errors. OpenAI says the evaluation does not represent typical usage and is designed around conversations that already contained a reported factual mistake.
OpenAI says GPT-6 Luna improves substantially as well. At higher effort settings, the company says Luna matches GPT-5.6 Sol on that evaluation at about one-hundredth of the cost.
The announcement does not provide a public error rate for the factuality test in the text describing the models. OpenAI says the evaluation is not controlled for response length, although its own verbosity tests found little dependence on answer length.
Availability begins outside standard ChatGPT
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 in the desktop app, while the models are not yet available in standard ChatGPT at launch.
Developers can call the models through the OpenAI API using the identifiers gpt-6-sol and gpt-6-luna. OpenAI says it will roll out the models gradually through ChatGPT during September 22 to manage service stability.
The pricing change is the most immediate distinction between the new models and their GPT-5.6 counterparts: Sol’s output price falls from $20 to $10 per million tokens, while Luna’s falls from $1.20 to $0.50, a 58.3% reduction. For developers running long coding sessions or high-volume agent workflows, those token rates—not the model names—will determine how much additional work fits inside the same budget.