Summary: On July 30, 2026, OpenAI cut GPT-5.6 Luna prices by 80% and Terra prices by 20%, while adding Fast mode for Sol at twice the standard price for up to 2.5 times faster processing. Luna falls to $0.20 per million input tokens and $1.20 per million output tokens; Terra falls to $2 and $12. The changes reflect claimed efficiency gains and expand low-cost options, but Sol remains preferable for ambiguity, uncertainty resolution, and high-stakes decisions.
The Changes: Lower Prices and a New Fast Mode
OpenAI used a July 30, 2026 announcement to cut prices on two GPT-5.6 models and add a new performance option for its flagship Sol model. The aim is to give developers more room to trade off cost, speed, and intelligence.
- GPT-5.6 Luna, the fastest and most affordable model, now costs 80% less. API pricing drops to $0.20 per million input tokens and $1.20 per million output tokens.
- GPT-5.6 Terra, the balanced model for everyday work, is now 20% cheaper at $2 per million input tokens and $12 per million output tokens.
- GPT-5.6 Sol keeps its existing pricing, but OpenAI introduced Fast mode in the API, which takes the place of the previous Priority Processing offering. Fast mode delivers up to 2.5× faster speeds than Standard processing at twice the price, with no change in intelligence. It is backward compatible: existing requests tagged
prioritywill automatically use Fast mode.
The same changes reach into how usage is counted in Codex and ChatGPT Work subscriptions. Terra and Luna usage now consumes fewer credits, but subscription prices and quota budgets remain unchanged. AWS pricing changes begin rolling out later on the same day.
The Efficiency Work Behind the Cuts
OpenAI credits these price cuts to improvements across three layers: the models themselves, the inference systems that run them, and the agentic harness that connects models to tools and context. GPT-5.6 models take a more direct path through work, better routing keeps hardware productive, optimized production software generates tokens more efficiently, and smarter context management helps agents avoid repeating completed work.
One example stands out. Inside a human-led process, GPT-5.6 Sol autonomously rewrote and optimized production kernels, designed and ran hundreds of experiments aimed at improving token generation, and monitored training, intervening when problems arose. That kernel work reduced the end-to-end cost of serving the model by 20%, while the experiments increased token-generation efficiency by more than 15%. OpenAI reads this as a tighter feedback loop: the more capable models become and the more autonomously they work, the faster efficiency gains arrive.
What This Means in Practice for Builders
OpenAI keeps returning to one principle: match intelligence to the outcome. How much intelligence, speed, reliability, and cost a task deserves depends on its stakes, the cost of error, its urgency, and its scale. That balance can shift from one step of a workflow to the next.
Take a coding workflow: it might use Sol to resolve uncertainty and define the plan, then use Luna to implement well-specified changes, write and run tests, and evaluate results. Every stage then gets the maximum useful intelligence while paying the right price for the value it creates.
Luna’s numbers deserve attention of their own. On professional work measured by Agents’ Last Exam, Luna outperforms Fable 5 at an estimated cost per task nearly 99% lower. It also delivers performance comparable to frontier-class models from a year ago at roughly 6 cents on the dollar per task and at nearly nine times the speed. That makes high-volume tasks like large-scale document analysis, customer-interaction classification, and routine implementation economical to run broadly.
Do not rush to switch everything to Luna, though. OpenAI advises defining the outcome and quality standard you need, then using evaluations to determine where additional intelligence materially improves results and where faster, lower-cost processing delivers the same quality. In steps requiring uncertainty resolution or strategic planning, Sol’s value remains irreplaceable.
Constraints and Trade-offs to Weigh
The price cuts are substantial, but a few trade-offs come with them:
- Luna and Terra are not Sol. Both are designed for high-volume, well-specified tasks. When reasoning gets complex, requirements are ambiguous, or decisions carry high stakes, Sol’s intelligence may be worth the premium.
- Fast mode doubles the price. It’s a speed option, not a discount. Reach for it only when response time is critical and the 2.5× speedup justifies the cost.
- Subscription credit changes mean that Terra and Luna usage consumes fewer credits while the subscription price itself is unchanged. That effectively hands subscribers more value, but it is not a direct price cut.
- The AWS rollout is gradual: pricing changes begin rolling out later on the same day, so availability may vary.
The Takeaway: Build for the New Cost Curve
OpenAI’s strategy is built for scale: investing in both more compute and more productive compute, matching each workload to the best-suited system. This supports both ends of the price-performance curve—low-cost Luna and Terra for high-volume work, and Fast mode for Sol when speed matters.
For product teams, this means AI applications that were previously cost-prohibitive may now be feasible. But it also requires building your own evaluation pipeline to know when to spend more for speed or intelligence. The price cuts are not a promotion; they reflect real improvements in model and infrastructure efficiency, and these improvements are likely to continue.
As OpenAI puts it, the gains can compound: more capable models help find the next generation of improvements, shortening the path to better performance and lower costs. What that asks of you: reassess your workflows, define quality standards, and use evaluations to decide where Luna, Terra, or Sol fits best. The frontier of affordable AI just moved.