OpenAI expands GPT-6 with Sol and Luna, cutting prices for production
OpenAI introduced GPT-6 Sol and Luna on October 7, extending frontier capabilities into cheaper, more deployable models. Developers now get lower API costs, stronger caching, and broader availability across ChatGPT Work, Codex, and the API.
OpenAI’s GPT-6 expansion shifts frontier capability into cheaper models
OpenAI introduced GPT-6 Sol and GPT-6 Luna on October 7, 2026, as a new expansion of the GPT-6 family. The practical importance is not just that OpenAI added more models, but that it pushed stronger capability into a family designed to be easier to deploy and cheaper to run.
According to OpenAI, Sol and Luna were trained with similar methods to GPT-6 Astra, bringing Astra-style gains in professional work, factuality, coding, computer use, and alignment into faster and more affordable models. OpenAI’s October 7 release notes also frame Astra as the higher-end model in the family, while Sol and Luna are intended to distribute frontier intelligence at lower cost.
What the price cuts mean for developers
OpenAI cut API prices for GPT-6 Sol and Luna by 50% versus GPT-5.6 promotional pricing. GPT-6 Sol is listed 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 listed 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.
That pricing shift matters because it makes stronger models more realistic for workloads that need repeated calls, tool use, or long-running workflows. In practice, lower token prices can change what teams are willing to move from testing into production, especially when the model is expected to handle coding, computer use, or workflow automation.
Benchmark claims point to a cheaper agentic option
OpenAI’s launch material emphasizes performance on agentic and software tasks. On AutomationBench at max effort, OpenAI says GPT-6 Sol outperforms Claude Opus 5 while costing 9% as much per task. OpenAI also says GPT-6 Sol scores 56.4% on Agents’ Last Exam at max effort, with 60% lower cost per task than Claude Opus 5.
The same release says GPT-6 Sol reaches 68.8% on DeepSWE v1.1 at max effort, within 1.1 percentage points of Claude Fable 5’s high score, at about 80% lower cost per task. On OSWorld 2.0 offline, OpenAI says GPT-6 Sol at xhigh effort scores 60.5%, nearly matching Claude Opus 5 at medium effort while costing about 80% less per task. For developers, the message is clear: OpenAI is positioning Sol as a lower-cost route to serious coding and computer-use work, not just a smaller convenience model.
Caching and availability may matter as much as the model cards
OpenAI says GPT-6 Sol and Luna now have improved prompt caching, with higher cache hit rates by default and 90% discounts on cached input-token reads. That can matter as much as raw model pricing, because repeated context is common in real applications and cached reads can substantially lower the cost of steady traffic.
OpenAI says GitHub’s use of these caching improvements has reduced the share of prompt tokens requiring fresh processing by more than 50% across billions of requests. The broader takeaway is that OpenAI is not only lowering list prices, but also improving the economics of repeated use, which is often what determines whether an application is affordable at scale.
Where Sol and Luna show up next
According to OpenAI, GPT-6 Sol and Luna are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, and in the API as gpt-6-sol and gpt-6-luna. OpenAI’s October 7 safety update says GPT-6 in ChatGPT replaces GPT-5.6 Sol and GPT-5.6 Luna in ChatGPT, while the versions in Codex and Work remain the earlier September releases.
The October 8 API changelog adds two deployment details worth watching. It says Ultrafast mode for GPT-6.1 Sol is available in the Responses API, and that GPT-6 Luna also appeared in the Decisions API beta on October 6. For teams planning rollout, the practical next step is to compare the new prices, caching behavior, and product availability against current production workloads, then test where lower-cost agentic work can move from experiment to default.
Sources
- Introducing GPT-6 Sol and Luna OpenAI · October 7, 2026
- GPT-6 Sol and GPT-6 Luna: October 2026 update OpenAI Deployment Safety Hub · October 7, 2026
- GPT-6 Astra: A new generation of intelligence OpenAI · October 7, 2026
- Changelog | OpenAI API OpenAI · October 8, 2026
This article was researched and drafted with AI from the sources above. Spot an error? Email info@hiddenproai.online.