OpenAI cuts Luna 80%, Terra 20% on GPT-5.6; launches Fast mode

On July 30, 2026, OpenAI announced significant pricing reductions and a new processing mode for its GPT-5.6 model family. Luna, the fastest and most affordable tier, now costs $0.20 per million input tokens and $1.20 per million output tokens—an 80% reduction that makes it one of the cheapest capable models on the market. Terra, the balanced everyday model, drops 20% to $2 per million input tokens and $12 per million output tokens. Sol, the flagship, retains its existing pricing.

The price reductions flow through OpenAI's subscription products as well. In Codex and ChatGPT Work, Terra and Luna now consume fewer credits against a user's quota, allowing the same subscription plan to stretch further without any change to subscription prices or quota budgets themselves.

OpenAI also introduced Fast mode for Sol, replacing the previous Priority Processing option. Fast mode delivers responses up to 2.5 times faster than Standard processing at twice the base price, with no change in model intelligence. The feature is backward compatible—existing API requests tagged as priority automatically route to Fast mode without requiring code rewrites. This aligns with a pattern Anthropic has already established with Claude Opus 4.8, which also runs roughly 2.5 times faster at twice the base rate.

The efficiency gains behind the price cuts stem from improvements across three layers: the models themselves, the inference systems serving them, and the agentic harness. Better routing keeps hardware busy, optimized serving software generates tokens more cheaply, and smarter context management prevents agents from repeating completed work. A particularly notable development is that GPT-5.6 Sol is now helping optimize itself. Within a human-led process, Sol autonomously rewrote and optimized production kernels and ran hundreds of experiments to improve token generation, reducing the end-to-end cost of serving the model by about 20% and increasing token-generation efficiency by more than 15%.

For developers, the broader implication is strategic. Rather than defaulting every call to the flagship Sol or assuming the cheapest tier is always right, teams can now match intelligence, speed, and cost to each workflow step. Luna is positioned for high-volume, cost-sensitive work; Terra for balanced quality and cost; and Sol for the most demanding reasoning tasks. On Agents' Last Exam, a benchmark for professional agentic work, OpenAI reports that Luna outperforms Fable 5 at an estimated cost per task nearly 99% lower. According to Artificial Analysis, Luna also sits well ahead of similarly priced models on its Intelligence Index. The practical pattern emerging is that cheaper Luna and Terra pricing makes it realistic to run agentic features at scale—where a coding agent, for example, might use Sol to resolve uncertainty and write a plan, then hand well-specified changes to Luna to implement, test, and evaluate.