On 30 July 2026, OpenAI drastically lowered the API prices of two models in the GPT-5.6 family, less than three weeks after their launch. Luna, the fastest and cheapest variant, fell 80% in price. Terra followed with a 20% reduction. That sounds like good news for users, and it is, but behind the discount lies an interesting shift in the AI market that you, as a professional, would do well to keep an eye on.
What has changed?
The GPT-5.6 family consists of three variants: Luna (fast, affordable), Terra (mid-range) and Sol (flagship). Prices per million tokens:
| Model | Input before | Input now | Output before | Output now |
|---|---|---|---|---|
| Luna | $1,00 | $0,20 | $6,00 | $1,20 |
| Terra | $2,50 | $2,00 | $15,00 | $12,00 |
| Sol | unchanged | unchanged | unchanged | unchanged |
Sol, the most capable model in the family, was left untouched. The discount therefore sits in the two lower tiers, precisely the models that compete most fiercely with open-weight alternatives such as Qwen, Mistral and Llama.
What is driving the discount?
OpenAI states that the price reduction is made possible by efficiency gains in its inference infrastructure, including using the model itself to optimise the production code. That may well be true, but the timing says something too. Introducing a steep discount three weeks after launch is unusual for a frontier lab. The pressure comes from open models that cost less and deliver comparable quality on many tasks. The incentive to stay with OpenAI is shifting from exclusive capacity to ecosystem and tooling.
For a comparison of models on price, quality and GDPR profile, including the recent GPT-5.6 variants, see our models page.
What does it mean for you?
Low-volume users (individual professionals, small teams): Luna at $0.20 per million input tokens is now negligibly cheap for most everyday tasks. One million tokens is roughly 750,000 words, more than enough for a week of intensive use.
High-volume users and tool builders: the 80% drop for Luna makes bulk processing of documents, emails and transcripts considerably more attractive via the API. The break-even point against open alternatives shifts, but it still exists, especially once you factor in hosting and maintenance costs.
Sectors with GDPR obligations: the price reduction changes nothing about the legal profile of the models. OpenAI processes API data through its EU region for Enterprise customers, but standard API contracts still warrant scrutiny. See our safety page for the current GDPR overview per model.
Broader context: prices are falling structurally
This is not the first time. In August 2026, the price of GPT-5.6 Luna also came down as part of a broader market trend. Earlier this year, Google too lowered Gemini prices with the introduction of Gemini 3.6 Flash (21 July). The pattern is clear: capacity at the top of the market (Sol, Claude Opus 5, Gemini Ultra) holds its price, while the middle tier keeps getting cheaper.
For professionals in sectors such as legal, finance or healthcare, this means API costs are less often the obstacle to adoption. The relevant questions shift to: which model fits your legal basis for processing, who manages it, and how do you document your decisions?
Sources: OpenAI price change 30 July, VentureBeat analysis, edtechinnovationhub details