OpenAI expanded its GPT-6 family on September 22 with two models cheaper and faster than the flagship GPT-6 Astra: Sol and Luna. The move landed just 90 minutes after Anthropic launched Opus 5.5 — the clearest sign yet that lab-to-lab competition is now being fought on per-token pricing as much as on benchmarks.
GPT-6 Sol is built for complex work like coding, while GPT-6 Luna targets high-volume, narrowly-scoped tasks: summarizing documents, classifying tickets, answering routine questions. OpenAI describes Luna as the "office workhorse" model — the one that churns through thousands of repetitive requests a day — and Sol as the one that now takes on tasks that used to require the flagship model.
According to OpenAI, gpt-6-sol became available via API that same day, September 22, with a 1,050,000-token context window and a 128,000-token maximum output. The rollout hit ChatGPT Work and Codex first for eligible paid plans, with gradual expansion to the app and website through launch day; Luna is also available on the desktop app and for users on the Free and Go plans.
OpenAI attributes the drop to improvements in caching and inference, not a simpler model. The company claims Sol makes "half the errors of its predecessor," reaching reliability close to Astra's at a much lower cost — and that both models outperform Anthropic's equivalent offerings on benchmarks, a claim from OpenAI that hasn't yet been verified by independent evaluations.
The pattern matters more than any single model: each new generation now ships a "premium" tier (Astra), a mid-priced workhorse tier (Sol), and a cheap high-volume tier (Luna). Before signing a contract or sizing infrastructure, it's worth mapping what percentage of a workflow's calls actually need the most expensive model — in most enterprise pipelines, it's a minority.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel