OpenAI cut the API price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% on July 30, 2026, in a repricing that takes its mid-tier model from $1.00 to $0.20 per million input tokens and from $6.00 to $1.20 per million output tokens. The changes applied immediately to all API accounts with no migration required, and the flagship GPT-5.6 Sol tier was left untouched at $5.00 input and $30.00 output.
The company framed the cut as an efficiency dividend rather than a discount, citing "internal efficiency improvements across model layers and inference systems" that let it complete more work on the same compute. The stated goal, in OpenAI's words, is to "push the price-performance frontier, making advanced AI more economical for a wider range of enterprise workloads."
Key Highlights
- GPT-5.6 Luna: $1.00 to $0.20 per million input tokens; $6.00 to $1.20 per million output tokens — an 80% reduction.
- GPT-5.6 Terra: $2.50 to $2.00 input; $15.00 to $12.00 output — a 20% reduction.
- GPT-5.6 Sol: unchanged at $5.00 input and $30.00 output.
- Fast mode replaces Priority Processing in the API. On Sol it delivers up to 2.5x faster responses at twice the standard per-token rate.
- ChatGPT and Codex subscription prices and quota budgets are unchanged, but Luna and Terra now consume fewer credits — the same weekly allowance covers more requests.
Details
The asymmetry of the cut is the story. OpenAI did not lower prices across the board; it repositioned one specific tier. Luna now sits back in the price band occupied by earlier model generations, while Sol gained a premium speed option instead of a discount. The result is a wider spread between the cheap workhorse tier and the frontier tier than OpenAI has previously maintained.
One detail matters for anyone running long-context jobs: when a single request exceeds 272,000 tokens, Luna doubles its input rate and applies a 1.5x multiplier to output charges. That surcharge changes the arithmetic considerably at the top end of the context window, and it is easy to miss when modelling costs from the headline rate alone.
Impact
For teams already shipping on the OpenAI API, the immediate effect is that workloads previously parked as too expensive become viable. "For CIOs, the biggest impact is likely to be scaling AI adoption rather than simply cutting costs," said Pareekh Jain of Pareekh Consulting, who expects inference costs to keep falling as new chips, better inference software and more efficient model designs compound.
Chandrika Dutt of Avasant expects enterprise teams to spend the savings rather than bank them, building "increasingly sophisticated agentic workflows that were previously difficult to justify economically." That is the Jevons paradox applied to tokens: cheaper units tend to expand consumption rather than shrink budgets. The practical advice from both analysts is the same — keep the architecture provider-flexible and treat AI FinOps as a standing discipline, because the pricing floor is still moving.
Background
The pressure is coming from open-weight models. DeepSeek V4 Flash undercuts the new Luna rate outright at $0.14 input and $0.28 output. DeepSeek V4 Pro trades places with Luna depending on the input-to-output ratio: for input-heavy work under 272,000 tokens Luna wins, but when input and output volumes are equal, a representative job costs $1.40 on Luna against $1.305 on V4 Pro.
The gap widens sharply on very long prompts. A single 900,000-token input request costs roughly $0.54 on Luna once the long-context surcharge applies, against about $0.154 on DeepSeek V4 Flash — a reversal of the advantage the new pricing was meant to establish.
That is the structural argument for why this cut happened. Open-weight releases including DeepSeek V4, Moonshot's Kimi K3 and Zhipu's GLM-5.2 have made a two-stage strategy routine: prototype on a commercial API, then move the high-volume production workload onto self-hosted infrastructure. Every price cut on the proprietary side is an attempt to make that second step less attractive.
For comparison, Anthropic's Claude Fable 5 lists at $10 per million input tokens and $50 per million output tokens — an order of magnitude above the new Luna rate, and a signal that the frontier tier and the volume tier are now priced as genuinely different products rather than points on one curve.
What's Next
OpenAI separately opened ChatGPT for Academic Researchers on July 29, extending free access to its most capable models — including GPT-5.6 Sol Pro — to 100,000 scientists, mathematicians and engineers. The rollout starts with 10,000 participants and expands through 2027, limited to recognised degree-granting universities with high research activity.
The open question is whether the cuts hold. Analysts polled by InfoWorld consider the reductions broadly sustainable, on the view that hardware and software efficiency gains will keep pushing cost per token down. The counter-argument, made in Forbes, is that a price war among foundation model providers looks like the early consolidation phase of an infrastructure industry — the same pattern that concentrated chipmaking and cloud into a handful of players. Either way, teams budgeting AI spend for the next two quarters should assume today's rate card is temporary.
Source: OpenAI