OpenAI cuts the price of GPT-5.6: what it changes

In late July 2026, OpenAI cut the price of two of its GPT-5.6 models by up to 80% — then, on 21 August, cut its top-end model in turn. If you use ChatGPT every day without ever touching an API, the news may seem technical and remote. It is not entirely: here is what it changes, and for whom.

What changed, concretely

OpenAI first cut, on 30 July 2026, the price of two of its three GPT-5.6 models — Terra and Luna — without touching the top-end one, Sol. Then, on 21 August, the company cut Sol in turn by more than 20% (a promotional cut, detailed below). The rate is expressed in dollars per million tokens (the base unit a language model processes), distinguishing the text sent to the model (“input”) from its answer (“output”). Here is the state of the rates in late August 2026:

Model	Current rate (input / output, per million tokens)	Latest cut
Sol (the most capable)	$4 / $20 (promo through 21 Nov 2026)	−20% input / −33% output, on 21 August
Terra (intermediate)	$2 / $12	−20%, on 30 July
Luna (the fastest)	$0.20 / $1.20	−80%, on 30 July

According to OpenAI, these cuts come from efficiency gains across the whole technical chain: optimised hardware routing, improved inference software, and context-caching algorithms that spare AI agents from redoing calculations already performed.

Why it concerns you, even without touching the API

You probably do not write code that calls GPT-5.6 directly. But a large share of the tools you already use — writing assistants, browser extensions, AI features built into business software — are themselves built on this kind of API. When the cost of the technical building block drops by 80% on the “fast” model, the makers of these tools gain a margin they can pass on as lower prices, broader free features, or more responsive tools.

It is also a market signal: such a price cut comes amid intense competition on the cost of AI, with similar announcements or competitive open models from Anthropic, Google and Chinese players like DeepSeek or Moonshot AI over the summer of 2026.

What it does not change

Your ChatGPT Plus or Team subscription is not directly affected: this is an API rate, aimed at developers and companies that build products on these models, not at people who use the consumer ChatGPT interface. If you are comparing ChatGPT, Claude and Gemini for your use today, the criterion that matters stays the same: what each tool does well for your tasks, not the per-token rate of a model you never call directly.

The 21 August Sol cut: a price war setting in

On 21 August 2026, OpenAI extended the move to its top-end model: the price of Sol drops from $5 to $4 per million input tokens (−20%) and from $30 to $20 per million output tokens (−33%). It is a promotional cut, announced through at least 21 November 2026, and it applies only to the pay-as-you-go API, Codex credits and eligible ChatGPT Work plans — not to Plus, Pro or Business subscriptions. The press reads it mainly as a move to slip under the rate of Claude Opus 5 on the most-capable segment: after months of holding its position, OpenAI is entering the price war on the “frontier” side in earnest.

Key takeaway

A price cut on an API is not a consumer event, but its effects spread through everyday tools: cheaper or more generous AI features at third-party makers, and pressure on the whole market that ultimately benefits those who compare and choose their tools methodically.