Advancing the price-performance frontier with GPT‑5.6
Points and comments are a snapshot, not live.
OpenAI cuts GPT-5.6 Luna prices 80% and Terra 20%, citing model-driven efficiency gains.
OpenAI announced price cuts for GPT-5.6 Luna (80% less) and Terra (20% less) starting July 30, driven by model and infrastructure efficiency improvements. GPT-5.6 Sol autonomously rewrote production kernels, cutting end-to-end serving costs 20%, and increased token-generation efficiency by 15%. A new Fast mode for Sol offers up to 2.5x faster speeds at double the price. Luna costs $0.20/M input tokens and $1.20/M output tokens; Terra costs $2/M input and $12/M output. Sol pricing remains unchanged. The changes apply to the API, ChatGPT Work, and Codex.
What commenters are saying
Commenters generally welcomed the cuts, seeing competitive pressure from Chinese models (DeepSeek, GLM, Kimi) as a driver. Some noted Luna's performance is close to Sol for many tasks, though others flagged task-dependent differences. A debate emerged on whether inference cost savings are significant given OpenAI's massive spending. One commenter argued electricity costs give China an advantage, but others countered that US industrial electricity is cheaper. Many saw the model-optimized kernel improvements as a notable technical milestone.