OpenAI has rolled out updates to its GPT-5.6 model line up, looking to cut costs and boost speed for both developers and businesses. The company is dropping prices for GPT-5.6 Luna and Terra, bringing in a new Fast mode for GPT-5.6 Sol through the API, and tweaking how usage is tallied in ChatGPT Work and Codex. The changes come from upgrades in how GPT-5.6 models are built and deployed. OpenAI says its systems are running more efficiently, which has lowered operational costs for the models. That, in turn, led to price reductions, though subscription fees for ChatGPT and Codex stay the same.
What OpenAI Changed for Luna, Terra, and Sol
Among the updates, the price cuts for Luna and Terra stand out. Luna now costs 80% less, while Terra is down by 20%. Under the new API pricing, a million input tokens for Luna run $0.20, output tokens are $1.20 per million. Terra, meanwhile, is priced at $2 per million input and $12 per million output tokens. Pricing for GPT-5.6 Sol has not changed. OpenAI is also swapping the Priority Processing option for a new Fast mode on GPT-5.6 Sol in the API. Fast mode delivers speeds up to 2.5 times faster than Standard processing, though it costs twice as much as Standard. API requests labelled as priority will automatically shift to Fast mode.
There is more: OpenAI is updating the Auto review feature in the ChatGPT app and Codex CLI, moving it from GPT-5.4 to GPT-5.6 Luna. Because of Luna’s new pricing, the company estimates a 10 times drop in the cost of Auto review. For users of ChatGPT Work and Codex, Luna and Terra now use fewer credits, stretching paid quotas further. Even with these changes, OpenAI says subscription prices and quota limits for ChatGPT and Codex are not changing.
The company also explained how these changes became possible. OpenAI said GPT-5.6 Sol helped improve production kernels, carried out hundreds of experiments to increase token generation efficiency, and monitored training processes. According to the company, this reduced the end to end cost of serving the model by 20% while increasing token generation efficiency by more than 15%. GPT-5.6 Terra and Luna continue to be available through ChatGPT Work, Codex, and the OpenAI API. Free and Go users get access to Terra, while Plus, Pro, Business, and Enterprise accounts can use both Terra and Luna.
We are committed to pushing the model frontier across cost efficiency, capability, and speed.
— OpenAI (@OpenAI) July 30, 2026
Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.
Luna and Terra’s lower prices are… pic.twitter.com/rFhK7XKedp
How GPT-5.6 Luna Compares to Fable 5 and Rivals
As part of the announcement, OpenAI included a comparison showing where it believes the GPT-5.6 family fits among competing AI models. They claim Luna delivers performance comparable to last year’s frontier models at just six cents on the dollar per task and runs almost nine times faster. They also say that, measured by Agents’ Last Exam, Luna beats Fable 5 for professional tasks while keeping estimated costs nearly 99% lower. That comparison is based on OpenAI’s own testing and aims to show how the GPT-5.6 models balance power, speed, and price.
OpenAI suggests businesses use Sol for more complex planning and reasoning, and rely on Luna for tasks like implementation, testing, and other well defined jobs that benefit from lower costs. Some of these models have also been put to the test by independent organizations. Artificial Analysis ranks GPT-5.6 Luna as a standout for its price, putting it close to the bigger frontier models on its Intelligence Index. Still, benchmark results are not all the same. On SWE Bench Pro, for instance, Claude models come out ahead of GPT-5.6 models, while other coding and agent benchmarks show smaller gaps. So OpenAI’s comparisons should be weighed alongside these outside evaluations, not taken as the final word.
Also read: OpenAI’s GPT-5.6 Turns AI Into a Three-Level Product Line with Sol, Terra, and Luna
