OpenAI, Anthropic Cut API Prices Up to 80%

OpenAI and Anthropic have cut the price of their API models by as much as 80% since late July, a run of reductions that data scientist Max Woolf laid out in an Oct. 25 analysis on his blog, minimaxir.com. The official pricing pages show the cheapest OpenAI tier now lists at one-tenth of its pre-July input rate, while Anthropic's newest Opus model costs 20% less per token than the one it replaced.

Key takeaways

• OpenAI cut GPT-5.6 Luna by 80% on July 30, then priced its successor, GPT-6 Luna, at $0.10 per million input tokens and $0.50 per million output tokens on Sept. 22.
• Anthropic launched Claude Opus 5.5 on Sept. 22 at $4 input, $20 output and $0.20 cache reads per million tokens, down from $5, $25 and $0.50 for Opus 5.
• Woolf argues cheap open-weight models such as DeepSeek V4.1 Flash, not goodwill, are setting the floor, and calls the idea of a coming price spike "more a fantasy."

How far OpenAI and Anthropic API prices have fallen

The first big move came on July 30, when OpenAI said GPT-5.6 Luna, "our fastest and most affordable model, will cost 80% less," at $0.20 per million input tokens and $1.20 per million output tokens. On Sept. 22 it replaced the model with GPT-6 Luna at $0.10 and $0.50, which OpenAI described as half the GPT-5.6 rate.

The mid-tier followed the same path. GPT-5.6 Sol was discounted to $4 input and $20 output, and GPT-6 Sol arrived on Sept. 22 at $2 and $10. A week later, GPT-6.1 Sol kept those rates but halved cached input to $0.10 per million tokens.

Anthropic's cuts have targeted caching and its flagship tiers. Claude Fable 5.1, released in September, kept $10 input and $50 output pricing, but "cache reads now cost 75% less, or $0.25 per million tokens," the company said. Opus 5.5 lowered every Opus rate, and Anthropic says it "will cost 40% less than Opus 5 on typical workloads."

Why cache read pricing matters for AI coding agents

Woolf's analysis puts the spotlight on cached tokens rather than headline rates. "When using agents for coding, cache read costs are the vast majority (96% - 99%) of the total token expenditure," he wrote, so a cut to cache reads moves a developer's bill almost one for one.

That is why changes such as Fable 5.1's 75% cache cut and GPT-6.1 Sol's halved cached input can matter more to agent-heavy teams than unchanged list prices suggest. Our earlier report on Claude Opus 5.5's 20% price cut covers that launch in detail.

Context: open-weight models set the floor

Woolf argues the cuts are not routine, writing that "no one cuts prices by 80% unless there is some externality in play." He points to DeepSeek, whose V4.1 Flash 1017 model he lists at $0.15 input, $0.60 output and $0.003 for cache reads, served by 29 distinct providers on OpenRouter.

With that many hosts competing on the same open weights, closed-model vendors face a ceiling on what they can charge for comparable work. Woolf also notes the timing is unusual, since both companies are reportedly preparing for public offerings, when investors watch losses closely.

What's next for AI API pricing

Woolf rejects the view that today's prices are subsidized and must jump later. "The idea that token prices will suddenly rise is more a fantasy," he wrote, arguing that competition pushes prices toward the marginal cost of serving a token.

For buyers, the practical takeaway is that a model's sticker price now changes within weeks, often with a new version number attached. Teams still on older models may find the newer tiers cheaper, but benchmarks and per-task costs should be checked before switching, since a lower token rate does not guarantee a lower bill for every workload.

FAQ

How much did OpenAI cut GPT-5.6 Luna's price?
By 80%, effective July 30, to $0.20 per million input tokens and $1.20 per million output tokens. Its successor, GPT-6 Luna, costs $0.10 and $0.50.

What does Claude Opus 5.5 cost through the API?
$4 per million input tokens, $20 per million output tokens and $0.20 per million cache-read tokens, compared with $5, $25 and $0.50 for Opus 5.

Will AI token prices go back up?
Nobody can say for sure. Woolf argues competition from open-weight models makes a sudden rise unlikely, but neither OpenAI nor Anthropic has committed to future pricing beyond its current published rates.

References

• OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why? — Max Woolf, minimaxir.com (Oct. 25, 2026)
• Advancing the price-performance frontier with GPT-5.6 — OpenAI (July 30, 2026)
• Introducing GPT-6 Sol and Luna — OpenAI (Sept. 22, 2026)
• Claude Opus 5.5 — Anthropic (Sept. 22, 2026)
• Claude Fable and Mythos 5.1 — Anthropic (September 2026)