🔍 Read the full analysis: The AI Price War: OpenAI And Anthropic Cut Prices As They Chase Each Other’s Customers on ThorstenMeyerAI.com
Get movie nights delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI and Anthropic have cut prices on several AI models in recent months, while reports describe additional OpenAI discounts aimed at major customers. The widely cited 80% to 90% savings figure describes one customer’s expected total cost reduction, not a discount from OpenAI’s list prices. Usage limits, credits and seat requirements are also changing, making total contract terms important alongside per-token prices.
OpenAI and Anthropic have lowered prices for several AI models since June, and a new report says OpenAI is offering some business customers additional discounts as the companies compete for major accounts. The changes could reduce costs for companies buying AI services, but reported savings vary by customer and do not represent a uniform cut in vendors’ public prices.
According to the vendors’ pricing pages, OpenAI’s Sol model prices moved from $5 per million input tokens and $30 per million output tokens in June to $2 and $10 for GPT-6.1 Sol in September. The smaller Luna model moved from $1 and $6 per million tokens to 10 cents and 50 cents. Those figures compare different model versions; they do not establish that an identical model was discounted by the same amount.
Anthropic also reduced prices across its model lineup. Its Opus price moved from $5 input and $25 output per million tokens for Opus 5 in July to $4 and $20 for Opus 5.5. Anthropic also cut the price of cached input for one model, Fable 5.1, and said that change lowers costs on typical workloads by about 25%, and by up to around 45% for highly agentic workloads. Those workload savings are the company’s estimates, not independent measurements.
The Information reported on October 24 that OpenAI has used customer-specific discounts to compete for large accounts. The source material says one offer to Replit included a discount of more than 30% on a contract worth over $100 million a year. DataCamp CEO Jonathan Cornelissen said OpenAI’s pricing could cut his company’s costs by 80% to 90%. That figure refers to expected overall savings against DataCamp’s existing setup, not a 90% reduction from OpenAI’s published list price.
Infographic 01 · The price slide
Four months, a fraction of the price.
What a million output tokens cost on each vendor’s price list, from the June previews to today. Bars show the price; shorter is cheaper.
OpenAI · output price per 1M tokens
List prices, standard tier
Anthropic · output price per 1M tokens
List prices, standard tier
Same scale in both panels. Fable 5.1 kept its $50 output price and cut cache reads instead.
The quieter cut: cache reads
matters most for agents
Cache-read price per 1M tokens, same scale in all three.
Woolf estimates cached tokens make up 96–99% of token spend in agentic coding. Anthropic says Opus 5.5 costs 40% less than Opus 5 and Fable 5.1 about 25% (up to about 45%) less than Fable 5 on typical workloads; those are vendor claims. Most rows compare a successor model with its predecessor. Anthropic’s newer tokenizer produces about 30% more tokens for the same text, so per-token prices are not directly comparable across vendors.
Sources: OpenAI API pricing (26 Oct 2026); OpenAI announcements (26 Jun–29 Sep 2026); Anthropic API pricing (26 Oct 2026); Anthropic announcements (30 Jun–22 Sep 2026); Vercel AI Gateway index (Oct 2026); Max Woolf (25 Oct 2026).
Powered by Thorsten Meyer AI
Lower Prices, More Complex Contracts
Falling public prices and private offers give companies a stronger basis for renegotiating AI contracts or testing more than one provider. But a lower token price alone may not determine what a customer ultimately pays. The cost depends on which models a company uses, how much input is cached, the volume and type of work, and any discounts or allowances included in its contract.
Contract terms can change even when a subscription’s monthly price does not. The source material says Anthropic added capped, expiring API credits to its Max and Team plans, while its Enterprise plan requires a 20-seat minimum. It also says OpenAI reduced the usage included with new $200 Pro subscriptions without changing the monthly fee. Buyers comparing offers need to check limits, eligibility, expiry dates and seat requirements, not only headline rates.
For employees and smaller organizations, the changes can affect which models are available within existing plans, but they do not guarantee a lower bill or more usage. A model’s successor may have different capabilities, and reported savings on one company’s workload may not transfer to another. The practical test is whether a price change lowers the cost of the specific tasks a customer needs to complete.
Popular right now
Dyson Ball Animal 3 Extra Upright Vacuum
Ninja Flip Toaster Oven and Air Fryer, 8-in-1, Stainless Steel, SP151
KitchenAid Classic Series 4.5 Quart Tilt-Head Stand Mixer K45SS, Onyx Black
As an affiliate, we earn on qualifying purchases.
How the Price Cuts Accumulated
The changes described in the source material took place over several months. OpenAI previewed GPT-5.6 Sol and Luna in June. Anthropic launched Sonnet 5 that month with introductory prices of $2 input and $10 output per million tokens, later making that pricing permanent. OpenAI cut Luna’s price in July and temporarily reduced Sol’s price in August before the September launches of newer models.
In September, OpenAI introduced GPT-6 Sol and Luna at lower prices than the earlier models, while Anthropic launched Opus 5.5 at lower listed rates than Opus 5. OpenAI then reduced GPT-6.1 Sol’s cache price on September 29. The source says data scientist Max Woolf tracked the sequence and that the figures were checked against company announcements and pricing pages retrieved on October 26.
Token rates are only one part of the comparison. Cached input prices can matter for software agents that repeatedly use the same conversation context. The source attributes to Woolf an estimate that cached tokens make up 96% to 99% of token spending in agentic coding. That estimate helps explain why cache pricing may affect some coding workloads more than the standard input rate, though the share will depend on how a particular system is used.
“I think ChatGPT is better now.”
— Amjad Masad, Replit CEO, speaking at TechCrunch Disrupt on October 16, as reported in the supplied source
What the Reported Savings Cover
The public information does not give a complete account of the terms attached to each private offer, including how long discounts last, what usage volumes qualify or whether other contract conditions apply. The Information’s report is described in the source material through its public summary and accounts by other outlets; readers should distinguish those reported terms from the vendors’ published prices.
The 80% to 90% cost figure is an estimate attributed to DataCamp’s CEO. The source does not provide the underlying calculations or enough detail about DataCamp’s previous spending to independently verify the expected reduction. Nor do the listed prices alone show whether a newer model performs as well as an older one on a particular customer’s tasks.
Some customer relationship details also remain unsettled. Replit’s CEO said the company was no longer working with Anthropic at an October event, then told TechCrunch that Replit remained model agnostic and described OpenAI as its preferred partner. A person familiar with the partnership told TechCrunch it was still in place. The supplied material does not resolve the scope or current status of that relationship.
Compare Usage and Contract Terms
Companies considering a change should compare providers using their own workloads and contract details, including input and output rates, cache pricing, usage allowances, credit expiry and minimum seat counts. Testing model quality on real tasks can help separate a lower published rate from a lower cost for the work the company actually needs.
Customers should also watch for further price and plan changes from both providers. The source material does not identify a confirmed next pricing announcement or a date for one. Until companies disclose more details about specific offers, private discounts and total customer savings will remain harder to compare than public model prices.
Key Questions
Did OpenAI cut every model’s price by 90%?
No. The 80% to 90% figure was DataCamp CEO Jonathan Cornelissen’s estimate of expected total savings against the company’s existing setup. It was not a 90% cut in OpenAI’s published prices.
Which OpenAI prices changed?
The supplied source says Sol moved from $5 input and $30 output per million tokens in June to $2 and $10 for GPT-6.1 Sol in September. Luna moved from $1 and $6 to 10 cents and 50 cents. These figures compare different versions of the models.
What did Anthropic change?
Anthropic lowered the listed rates for Opus 5.5 compared with Opus 5, and reduced cache-read pricing for Fable 5.1. The company’s reported estimates of workload savings depend on how the models are used.
Why do plan limits matter if prices are falling?
Usage allowances, credit caps, expiry dates and seat minimums can affect a customer’s total cost even when a monthly subscription fee stays the same. Buyers need to compare the full contract terms, not only the listed model price.
Source: ThorstenMeyerAI.com
