OpenAI launches GPT-6 Sol and Luna at half the price of GPT-5.6

OpenAI has filled in the GPT-6 family below Astra with GPT-6 Sol and GPT-6 Luna, "trained with similar methods as GPT-6 Astra". Sol is pitched at complex coding and professional work, Luna at fast, high-volume jobs like summarization. The headline is price: $2/$10 per million input/output tokens for Sol and $0.10/$0.50 for Luna, 50% below GPT-5.6's promotional rates. OpenAI credits more efficient caching and inference, and says the cut is permanent.

The GPT-6 lineup: Astra at $10/$50, Sol at $2/$10, Luna at $0.10/$0.50 — @OpenAI

OpenAI's benchmark numbers are framed almost entirely as cost per task against Claude. On AutomationBench, Sol (xhigh) scores 33.2% at $0.27 a task, ahead of Astra (low) at 30.3% for 3.9x the cost, Claude Opus 5 (max) at 26.9% for 11.1x, and Fable 5.1 with an Opus 5 fallback at 31.4% for more than 8.9x. On Agents' Last Exam, Sol (max) reaches 56.4%, above Opus 5's best at 60% lower cost. On DeepSWE v1.1, Sol (max) scores 68.8%, 1.1 points behind Fable 5 (xhigh) at roughly 80% less, and Luna (max) 66.6%. OpenAI also says Sol makes about half as many factual mistakes as GPT-5.6 Sol on its internal eval.

AutomationBench: each GPT-6 model sits up and to the left of its GPT-5.6 predecessor — OpenAI, via @OpenRouter

Every comparison is vendor-reported, with vendor-chosen effort levels, and Fable 5 stands in wherever Fable 5.1 scores weren't available. The independent read is more modest: Artificial Analysis finds Intelligence Index and Coding Agent Index scores level with GPT-5.6 (Sol 48 vs 47, Luna 37 vs 37), with gains on some evals and regressions on others. What moves is cost. Both models shift the cost-efficiency frontier by roughly halving the price of the same intelligence.

Same score, half the cost: GPT-6 Sol and Luna against their GPT-5.6 predecessors — @ArtificialAnlys

OpenAI also improved prompt caching, with higher default hit rates and a 90% discount on cached input reads. GitHub reports the change cut the share of prompt tokens needing fresh processing by more than half across billions of requests.

Both models are rolling out 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. Free and Go users can try Luna in the desktop app. They are also live on OpenRouter and in Visual Studio.

#Why it matters

This is a price move more than a capability jump. Holding GPT-5.6-level intelligence while halving the price, and making the cut permanent rather than promotional, resets what a mid-tier agentic workload costs. The Claude comparisons are aimed squarely at Anthropic, whose Opus 5.5 launched the same day at $4/$20 with its own claim of 40% lower running costs.

Sources: OpenAI announcement · @OpenAI launch post · @OpenAI rollout · Artificial Analysis · OpenRouter