🔍 Read the full analysis: Before You Sign Up: OpenAI And Anthropic’s AI Price War, Explained on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI and Anthropic have reduced prices for several newer AI models, while reporting describes additional discounts and changing customer usage. A widely cited 80% to 90% savings figure refers to DataCamp’s expected total cost reduction, not a discount of that size on OpenAI’s list prices.
OpenAI and Anthropic have cut prices for several AI models since June and are also offering or adjusting customer terms, according to their pricing information and reporting published this weekend. The changes could lower costs for companies buying AI services, but the headline savings figures do not all describe list-price cuts, and plan limits and contract conditions matter alongside token rates.
OpenAI’s pricing pages show its Sol model moving from $5 input and $30 output per million tokens in June to $2 and $10 for GPT-6.1 Sol in September. Its smaller Luna model moved from $1 and $6 to $0.10 and $0.50. These figures compare different model releases, rather than showing a price change to an unchanged model. The source says it checked the figures against vendor announcements and pricing pages retrieved October 26.
Anthropic also reduced prices across newer offerings. Its Opus line 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 cache-read pricing for Fable 5.1, while saying that caching can reduce costs on certain workloads. Those estimates are Anthropic’s own claims, not independent measurements. The source material does not provide enough detail to compare performance across all the model versions.
Separately, The Information reported on October 24 that OpenAI is offering discounts to attract customers using Anthropic. The reported deals include a discount of more than 30% for Replit and sub-$7 seats for a U.S. bank. Replit CEO Amjad Masad said at an October 16 event that the company was no longer working with Anthropic and that he thought ChatGPT was better; he later told TechCrunch that OpenAI was its preferred partner and that Replit remained model agnostic. A person familiar with the partnership told TechCrunch it was still in place.
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 Rates, Different Buying Terms
Lower per-token prices may reduce bills for developers and companies that make heavy use of AI, but the listed rate is only part of total cost. How much a customer pays also depends on which model is used, the volume and pattern of requests, cache pricing, included usage, and negotiated contract terms. A cheaper successor may not perform the same way on a buyer’s own tasks, so price comparisons alone cannot establish equivalent value.
The reported customer shifts suggest that pricing can influence which provider handles work. Rippling CEO Parker Conrad told The Information that OpenAI’s share of the company’s coding work rose from 16% in May to 52%, while Anthropic’s fell from 57% to 18%. Those figures describe one company’s use, not a market-wide trend. Still, if similar changes occur elsewhere, they could affect vendors’ revenues and give customers more leverage in negotiations.
Contract details also can offset headline savings. The source reports that Anthropic added capped, expiring API credits to its Max and Team plans and set a 20-seat minimum for Enterprise. It says OpenAI reduced the usage included in new $200 Pro subscriptions without changing the subscription price. Those terms can matter to small firms and individual users as much as the advertised rate does.
Popular right now
TCL 85 Designer Series QD-Mini LED TV (Model 85A400U) with Google Smart TV
Shark Navigator Robot Vacuum with Self-Empty Base, Dark Grey (RV2120AE)
iRobot Roomba 105 Vac Robot Vacuum
As an affiliate, we earn on qualifying purchases.
How Prices Shifted Since June
The pattern developed over several months, with changes to both models and rates. OpenAI cut Luna pricing in July and reduced Sol pricing in August, before introducing lower-priced GPT-6 Sol and Luna models in September. Anthropic launched Sonnet 5 in June at introductory pricing and made that pricing permanent in August, then released Opus 5 in July and Opus 5.5 in September. These are successive model releases and pricing decisions, not a single across-the-board reduction to every service.
Data scientist Max Woolf published a timeline of price changes on October 25. The source says its author checked the listed prices against vendor announcements and current pricing pages. One detail in the timeline is cache pricing: cached input can be relevant for AI agents that repeatedly process conversation history. Woolf estimates that cached tokens can account for 96% to 99% of token spending in agentic coding, but that is his estimate. Anthropic separately says Fable 5.1 can cost around 25% less on typical workloads and up to around 45% less on highly agentic workloads.
Those workload estimates help explain why a headline input or output price may not predict a customer’s bill. They also should not be read as a guaranteed saving: the result depends on how a system is used. As with model performance, companies need to test costs against their actual workloads before treating vendor estimates as their own expected savings.
“I think ChatGPT is better now.”
— Replit CEO Amjad Masad, speaking at TechCrunch Disrupt
Deal Savings and Plan Limits
The reported deals are not a full public price schedule. The source material does not specify every contract’s duration, eligibility, included usage, or other conditions, making it difficult to compare negotiated rates directly with list prices. The Information’s article is paywalled, and the source account relies in part on its public summary and subsequent reporting by The Decoder and all-ai.de.
The widely repeated 80% to 90% figure is especially easy to misread: it is DataCamp’s CEO describing expected savings against his existing setup, not a stated discount from OpenAI’s standard rates. The source also does not provide enough detail to independently verify the projected savings or assess how they would change over time.
Other questions remain open. Replit’s public statements and the account from a person familiar with its partnership indicate that the relationship with Anthropic may not be fully described by a simple switch from one provider to another. The details of the U.S. bank seat offer and the extent of any customer-specific OpenAI discounts are also not fully disclosed. Neither company’s future pricing and plan allowances are established by the current reports.
Compare Contracts, Not Just Rates
For companies considering a deal, the next step is to compare vendors using the same workload and accounting for input, output, caching, and usage limits. Buyers can ask for the full terms of any discount, including how long it lasts, what usage it covers, and whether credits expire. They can also test successor models on their own tasks before assuming that a lower price means equivalent performance.
For individual subscribers, the key check is whether the plan’s included usage has changed even if the monthly fee has not. The source advises companies to negotiate and avoid relying on a single vendor where practical. Further pricing announcements, plan changes, and customer disclosures will show whether the current reductions persist and how broadly negotiated offers are being made.
Key Questions
Did OpenAI cut prices by 80% to 90%?
Not according to the figures provided. DataCamp CEO Jonathan Cornelissen described expected total cost savings of 80% to 90% against his company’s existing setup. That is different from an 80% to 90% reduction in OpenAI’s published prices.
How much did OpenAI’s listed model prices fall?
The source compares June and September prices for different model releases. Sol went from $5 input and $30 output per million tokens to $2 and $10 for GPT-6.1 Sol; Luna went from $1 and $6 to $0.10 and $0.50. These comparisons do not establish that the newer models perform identically to their predecessors.
Did Replit leave Anthropic for OpenAI?
Replit CEO Amjad Masad said OpenAI was the company’s preferred partner and that Replit remained model agnostic. A person familiar with the partnership told TechCrunch it was still in place, so the public information does not establish a complete break with Anthropic.
What should AI buyers check beyond token prices?
Check included usage, seat minimums, credit caps and expiry dates, cache rates, contract length, and whether the model meets the needs of the workload. Compare actual bills and performance in tests that reflect your own use.
Source: ThorstenMeyerAI.com
