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

OpenAI introduced GPT-6 Sol and GPT-6 Luna on September 22, 2026, expanding the GPT-6 model family it launched with GPT-6 Astra earlier in September and cutting API prices for both new models by 50% from their GPT-5.6 promotional pricing.

In its announcement, OpenAI said it trained the two models with methods similar to those behind GPT-6 Astra, extending the advances behind Astra’s performance in professional work, factuality, coding, computer use, and alignment to faster, more affordable models. OpenAI said Astra remains its best model across the board for the most demanding projects, and described Sol and Luna as advancing cost efficiency across the family.

Pricing and Model Roles

GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, down from $4 and $20 for GPT-5.6 Sol. GPT-6 Luna is priced at $0.10 per million input tokens and $0.50 per million output tokens, down from $0.20 and $1.20 for GPT-5.6 Luna. OpenAI described each change as a 50% reduction against GPT-5.6 promotional pricing and attributed the lower prices to improvements in caching and inference, saying it is passing the savings directly to users and customers.

OpenAI’s API documentation for GPT-6 Sol describes the model as built for complex coding and agentic workflows, while the GPT-6 Luna documentation describes Luna as OpenAI’s most efficient model for focused, high-volume tasks. Both models list a 1,050,000-token context window and a 128,000-token maximum output, accept text and image input with text output, and support reasoning effort settings from none through max, with medium as the default. Sol carries an April 20, 2026 knowledge cutoff and Luna a May 18, 2026 cutoff. Cached input is priced at 10% of the uncached rate, at $0.20 per million tokens for Sol and $0.01 for Luna, and cache writes are billed at 1.25 times the uncached input rate.

Reported Evaluation Results

On AutomationBench, which tests AI agents on end-to-end business workflows using 47 tools, OpenAI reports GPT-6 Sol at xhigh effort scored 33.2% at $0.27 per task, outperforming Claude Opus 5 at max effort at 9% of that model’s cost per task and exceeding both low-effort GPT-6 Astra at 30.3% and Claude Fable 5.1 with Opus 5 fallback at 31.4%. OpenAI noted that the Fable 5.1 figure understates actual cost because it omits the fallback spending, which occurred on roughly 40% of tasks.

OpenAI reports GPT-6 Luna at high effort improves on its GPT-5.6 predecessor by 5.4 percentage points at 58% lower cost per task on the same benchmark. On Agents’ Last Exam, which evaluates long-horizon professional tasks across 55 sub-industries, OpenAI reports GPT-6 Sol at max effort scored 56.4%, above Claude Opus 5’s highest score in the evaluation at 60% lower cost per task.

On an internal factuality evaluation built from de-identified conversations in which users had flagged model mistakes, OpenAI said GPT-6 Sol makes about half as many mistakes as its predecessor and GPT-6 Luna at higher effort levels matches GPT-5.6 Sol at about a hundredth of the cost. OpenAI noted the flagged conversations are not representative of typical usage, where it said factual errors are rarer.

On FrontierCode, which grades coding agents on whether their changes are ready to merge into real codebases, OpenAI reports GPT-6 Sol improves substantially over GPT-5.6 Sol and matches Claude Fable 5.1 at xhigh effort at much lower cost. On DeepSWE v1.1, which tests complex software-engineering tasks in real codebases, OpenAI reports GPT-6 Sol at max effort scored 68.8%, within 1.1 percentage points of Claude Fable 5’s highest score in the evaluation of 69.9% at xhigh effort, at approximately 80% lower cost per task. OpenAI reports GPT-6 Luna at max effort scored 66.6%, comparable to Claude Opus 5 and Fable 5 at medium effort while costing 93% and 96% less per task, respectively.

OpenAI said its internal coding-agent usage has grown sharply, with daily token usage valued at API prices exceeding $600 for the median researcher and $7,000 for researchers at the 90th percentile.

On OSWorld 2.0 offline, OpenAI reports GPT-6 Sol at xhigh effort scored 60.5% against Claude Opus 5’s 60.3% at medium effort, at approximately 80% lower cost per task, and that GPT-6 Luna at max effort exceeded GPT-5.6 Sol at medium effort at one tenth of the cost.

OpenAI said evaluations of its own models ran in its research environment or through its API, that competitor scores came from publicly available reports, and that Claude Fable 5 scores were used where Fable 5.1 scores were unavailable. The company also said it carried Astra’s communication style into Sol and Luna, pointing to clearer phrasing, less jargon, and slightly shorter answers, particularly in technical and coding conversations.

Caching, Alignment, and Availability

OpenAI said improved prompt caching for GPT-6 delivers higher cache hit rates by default, with discounts of 90% on cached input-token reads. New developer tooling includes a Prompt Caching Dashboard, a diagnostics tool that explains missed caching opportunities, reasoning-effort and tool controls that preserve earlier context for cache reuse, and explicit breakpoints that let developers choose where cached prompt prefixes end. According to the announcement, GitHub reported that over the past several months the improvements reduced the share of prompt tokens requiring fresh processing by more than 50% across billions of requests to OpenAI models, helping Copilot respond faster.

On alignment, OpenAI said both models show improvements over their GPT-5.6 counterparts in its evaluations, including lower rates of misleading claims about their coding work, while noting the evaluations deliberately test challenging situations and do not measure failure rates in typical use. The full results appear in the GPT-6 Astra system card, according to the announcement.

GPT-6 Sol and Luna are available in ChatGPT Work and Codex starting September 22, 2026 for Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access GPT-6 Luna in the desktop app, and the models are not yet available in Chat. In the OpenAI API, the models are available as gpt-6-sol and gpt-6-luna. OpenAI said it plans to roll out the models in ChatGPT gradually throughout the day to keep service stable.