On July 30, 2026, OpenAI lowered API prices for the entire GPT-5.6 series. The smallest model was affected the most: GPT-5.6 Luna now costs 80 percent less, while the mid-range model, Terra, is 20 percent cheaper. In addition, OpenAI is introducing a new “Fast Mode” for the API, which replaces the previous Priority Processing for the top-tier Sol model. Subscription prices for ChatGPT and Codex remain unchanged—there, Terra and Luna now consume even fewer credits.
Key Points at a Glance
- Luna −80%: now $0.20 per million input tokens and $1.20 per million output tokens.
- Terra −20%: now $2 per million input tokens and $12 per million output tokens.
- Sol: Price unchanged, but with a new Fast Mode (up to 2.5× faster, double the price).
- Effective as of: July 30, 2026; rollout on AWS will follow shortly thereafter.
- Subscriptions: ChatGPT and Codex prices remain the same; Terra and Luna now cost fewer credits.
New API Pricing for GPT-5.6 Luna and Terra
OpenAI attributes the price reduction to efficiency gains across the entire model chain. Because GPT-5.6 is cheaper to operate, the company is passing the savings directly on to its customers. For anyone developing via the API, this primarily affects billing for large volumes. Here’s an overview of the new prices:
| Model | Input (per 1 million tokens) | Output (per 1 million tokens) | Change |
|---|---|---|---|
| GPT-5.6 Luna | $0.20 | $1.20 | −80% |
| GPT-5.6 Terra | $2.00 | $12.00 | −20% |
| GPT-5.6 Sol | unchanged | unchanged | New Fast Mode |
Luna is the fastest and most affordable model in the lineup and, according to OpenAI, is primarily intended for high-volume tasks: classifying documents, sorting customer inquiries, and implementing clearly defined code changes. The model can use tools and process multi-step workflows, which is why OpenAI positions it as a practical building block for agent-based applications. Terra remains the well-rounded model for everyday use, while Sol is the most powerful model for complex tasks.
Fast Mode Replaces Priority Processing
Along with the price reduction, OpenAI is introducing Fast Mode in the API. It replaces the previous Priority Processing tier and is aimed at anyone for whom response time is more important than price when using Sol. Specifically, GPT-5.6 Sol in Fast Mode operates up to 2.5 times faster than in standard processing, but costs twice as much. According to OpenAI, the quality of the responses remains unchanged.
Handy tip: If you’re already sending requests marked “priority,” you don’t need to change anything. These requests will continue to run automatically in Fast Mode. In Codex, this corresponds to the /fast command.
Just how affordable is GPT-5.6 Luna, really?
To put the price-performance ratio into perspective, OpenAI has provided benchmark data—which, as is always the case with manufacturer figures, should be taken with a grain of salt. According to the company, Luna now matches the performance of models that were considered top-of-the-line about a year ago, while costing about six percent as much per task and running nearly nine times faster.
The direct comparison with the competition is particularly striking: In the Agents’ Last Exam benchmark, which focuses on professional tasks, Luna is said to outperform Anthropic’s Claude Fable 5 model—at an estimated cost per task that is nearly 99 percent lower. In Artificial Analysis’s v4.1 index, Luna scores over 51 points at about $0.05 per task. According to OpenAI, this puts the model on par with or ahead of competitors such as Claude Opus 5 Low and Gemini 3.6 Flash, which are said to cost five to ten times as much per task.
Such comparisons come from the provider itself and are based on selected benchmarks. They provide a general indication but are no substitute for conducting your own tests with the relevant tasks. Anyone weighing their options among AI providers should also keep an eye on the current data center initiative by Anthropic and AMD, as more affordable infrastructure will eventually lead to lower model prices.
Sol has contributed to its own optimization
OpenAI attributes the efficiency gains to improvements in three areas: model architecture, inference systems, and the agent-based runtime environment that connects the models with tools and context. The most notable point: GPT-5.6 Sol is said to have been actively involved in identifying these optimizations.
According to OpenAI, within a human-supervised process, Sol independently rewrote production kernels, designed and conducted hundreds of experiments to improve token generation, and monitored training while intervening when problems arose. The kernel work reduced end-to-end costs for model operation by 20 percent, while the experiments increased the efficiency of token generation by more than 15 percent. OpenAI describes this as a feedback loop: The more autonomously the models operate, the faster additional efficiency gains can be identified. This concept of self-improving AI is currently driving half the industry—similar to ASUS’s agent-based AI software, Zenni Claw, the trend is clearly moving toward systems that process tasks independently.
What’s changing for ChatGPT and Codex users?
For now, subscription prices will remain the same. The plans and usage limits for ChatGPT and Codex will not change. However, users of Terra or Luna within a paid subscription will consume fewer credits—so their existing usage allowance will last longer. In ChatGPT Work and Codex, Free and Go users have access to Terra, while Plus, Pro, Business, and Enterprise users can choose between Terra and Luna. The model portfolio thus remains tiered: affordable and fast with Luna, balanced with Terra, and maximum performance with Sol. Google’s integration of Gemini Omni into the video editor Vids, for example, illustrates how other major AI providers are positioning themselves.
Frequently Asked Questions About GPT-5.6 and the Price Reduction
How much cheaper will GPT-5.6 be?
GPT-5.6 Luna costs 80 percent less, and GPT-5.6 Terra costs 20 percent less. The price for the top-of-the-line Sol model remains unchanged.
How much does GPT-5.6 Luna cost now?
As of July 30, 2026, Luna costs $0.20 per million input tokens and $1.20 per million output tokens in the API.
What is the new Fast Mode?
Fast Mode replaces the previous Priority Processing in the API. With GPT-5.6 Sol, it runs up to 2.5 times faster than standard processing, but costs twice as much. The quality of the output remains the same.
Will ChatGPT subscriptions get cheaper now?
No. The subscription prices for ChatGPT and Codex remain unchanged. However, Terra and Luna use fewer credits within the plans, so your existing credit allowance will last longer.
Conclusion: The price war over AI models is entering the next round
An 80 percent drop in price sounds spectacular, but it’s primarily another step in a trend that’s shaping the entire industry: AI inference is getting cheaper month by month. This is good news for developers because it makes applications viable that would have been too expensive just a year ago—especially in the field of agent-based systems, where many small model calls are aggregated.
There are two things to keep in mind. First, OpenAI’s benchmark figures are provided by OpenAI itself and tailored for promotional comparisons; whether Luna truly measures up to Claude or Gemini in your own use case is something you’ll have to determine through your own testing. Second, while OpenAI is lowering API prices, it’s leaving subscription costs for end users unchanged—for now, the price war is playing out in the token-based market. It will be interesting to see how Anthropic and Google respond. The race for the best value for money is likely to bring a few more announcements like this in 2026.
