GPT-6 Sol and Luna
Points and comments are a snapshot, not live.
OpenAI releases GPT-6 Sol and Luna, halving prices versus GPT-5.6 while improving performance.
OpenAI launched GPT-6 Sol and Luna, cheaper siblings of GPT-6 Astra, with API prices cut 50% versus GPT-5.6 equivalents (Sol: $2/$10 per 1M tokens; Luna: $0.10/$0.50). Sol outperforms Claude Opus 5 on AutomationBench at 9% of its cost per task, and scores 56.4% on Agents' Last Exam. Factuality errors are halved versus GPT-5.6 Sol. Coding benchmarks show Sol matching Claude Fable 5.1 xhigh at much lower cost. Prompt caching improvements reduce fresh processing tokens by over 50% for GitHub Copilot. Models are available today in ChatGPT Work, Codex, and the API.
Both models include alignment improvements and lower deception rates in coding evaluations.
What commenters are saying
Commenters largely welcome the price cuts, calling Luna pricing "insane" and noting it enables long-running agentic tasks without hitting limits. Several debate whether the cuts are a response to competition from Qwen4, GLM, Anthropic, or internal efficiency gains. Some users prefer GPT-6 Sol over Opus 5.5 on value, while others report still favoring Claude for code quality. A few express skepticism about OpenAI's data privacy and note the need for Azure/AWS availability (already confirmed in replies).