GPT-5.6 Luna Price Cut: What OpenAI's 80% Drop Means

On July 30, 2026, OpenAI cut the price of GPT-5.6 Luna by about 80% and Terra by 20%. Luna input fell from $1.00 to $0.20 per million tokens. Luna output fell from $6.00 to $1.20. The flagship, Sol, did not change.

OpenAI framed the move as advancing the price-performance frontier. It says Luna now matches models that were frontier-class a year ago for roughly 6 cents on the dollar per task, at nearly 9x the speed.

This page explains exactly what changed, why it happened, and which high-volume workloads the new prices make affordable. It is written for buyers deciding what to do next.

What Changed on July 30, 2026

OpenAI cut GPT-5.6 Luna by about 80% and GPT-5.6 Terra by about 20%, effective July 30, 2026. The flagship Sol tier was left unchanged.

Luna input dropped from $1.00 to $0.20 per million tokens. Luna output dropped from $6.00 to $1.20. That is the largest single price move in the GPT-5.6 family so far.

Terra input fell from $2.50 to $2.00, and Terra output fell from $15.00 to $12.00. Sol held at $5.00 input and $30.00 output.

• Luna: the fastest, cheapest tier — now 80% cheaper.
• Terra: the balanced tier — now 20% cheaper.
• Sol: the flagship tier — no change.

Before and After: The New Cost Table

Here are the old and new prices side by side, in US dollars per million tokens. Use it to recheck any budget you built before July 30, 2026.

The numbers below come from OpenAI's own pricing page and its price-performance post. Only Luna and Terra moved.

For a typical job, output tokens usually cost more than input tokens, so the output cuts matter most for busy pipelines.

• Sol — Input: $5.00 → $5.00 (unchanged). Output: $30.00 → $30.00 (unchanged).
• Terra — Input: $2.50 → $2.00. Output: $15.00 → $12.00.
• Luna — Input: $1.00 → $0.20. Output: $6.00 → $1.20.
• Luna output is now one-fifth of its old price.

Why OpenAI Cut the Price

OpenAI cut Luna and Terra to stay ahead in a fast-moving price war for cheap, capable models. Rivals have pushed low-cost frontier tiers hard all year.

The pressure comes from cheap tiers like Gemini Flash and Claude Haiku, plus open-weight models such as GLM and DeepSeek. These options made high-volume work cheaper than Luna was.

OpenAI calls this advancing the price-performance frontier. Its argument is that a task on Luna now costs a small fraction of what the same quality cost a year ago.

• Competitive pressure from Gemini Flash and Claude Haiku.
• Open-weight models like GLM and DeepSeek pushing prices down.
• OpenAI's goal: keep the cheapest quality tier on its own platform.

OpenAI's Price-Performance Claims

OpenAI says Luna now delivers year-ago frontier quality for roughly 6 cents on the dollar per task, at nearly 9x the speed. These are OpenAI's own figures.

On Agents' Last Exam, OpenAI reports that Luna outperforms Claude Fable 5 at an estimated cost per task nearly 99% lower. That is a vendor benchmark, so read it with care.

We treat these as claims from the maker, not independent results. They are useful for direction, but you should test Luna on your own tasks before trusting the numbers.

• ~6 cents on the dollar per task versus year-ago frontier models.
• ~9x faster than that older frontier class.
• Beats Claude Fable 5 on Agents' Last Exam at ~99% lower cost per task (OpenAI's figure).

Which Workloads This Unlocks

The cut makes high-volume, repetitive AI work economical that was borderline before. When each call costs less, you can run far more of them.

Luna is built for speed and scale: summarization, drafting, classification, and routine automation. At $0.20 input and $1.20 output, these jobs get much cheaper at volume.

It also lowers the cost of cheap agents that make many small model calls. Multi-step agents add up fast, so a lower per-token price changes what is affordable.

• High-volume text classification and tagging.
• Bulk drafting of emails, product copy, and replies.
• Summarizing large document sets and transcripts.
• Routine automation and cheap multi-step agents.

Who This Is For and What to Do Now

This price cut is aimed at teams running large volumes of routine AI work, plus builders shipping agents and automations at scale. If you send many small calls, you are the target buyer.

First, reprice your existing Luna and Terra pipelines using the new numbers. A workflow that looked too expensive may now pay for itself.

Second, test before you switch. Move a small slice of a low-risk workload to Luna, measure quality against your current model, and only then scale up.

• Reprice current pipelines with the July 30 numbers.
• Pilot Luna on a low-risk, high-volume task first.
• Compare quality against your current model, not just cost.
• Keep Sol for the hardest reasoning — its price did not move.

The Honest Caveat: Sol Was Not Cut

The flagship Sol tier did not get cheaper. It held at $5.00 input and $30.00 output per million tokens.

So the headline savings apply to lighter work, not the hardest reasoning. If your task needs Sol's max or ultra reasoning modes, your cost is unchanged.

This matters for planning. Do not assume the whole GPT-5.6 family got cheaper — only Luna and Terra did.

• Sol pricing is unchanged after July 30, 2026.
• Complex coding, science, and deep reasoning still cost the same on Sol.
• Savings are concentrated in high-volume, lower-complexity work.