🔍 Read the full analysis: The AI Price War Explained: OpenAI And Anthropic’s Offers And Tradeoffs on ThorstenMeyerAI.com
Get hardware and tech essentials 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 several model prices since June 2026, and reporting points to OpenAI offering discounts to win enterprise customers. A widely cited 80% to 90% savings figure describes one customer’s expected total cost reduction, not a standard list-price discount. Seat minimums, usage limits and credits also affect what buyers actually pay.
OpenAI and Anthropic have cut prices for several AI models since June, while OpenAI is also offering reported discounts to attract business customers from Anthropic. The change is giving buyers more leverage, but the widely repeated claim of 80% to 90% savings refers to one customer’s expected overall cost reduction—not a standard reduction in OpenAI’s list prices.
Pricing figures compiled by AI pricing tracker Max Woolf and checked against vendor announcements and pricing pages on October 26 show substantial changes over four months. OpenAI’s Sol model went 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 are prices for different model versions, not necessarily reductions to the cost of using the same model.
Anthropic also reduced prices and changed how some usage is billed. Its Opus line went from $5 input and $25 output per million tokens for Opus 5 in July to $4 and $20 for Opus 5.5. Anthropic said Opus 5.5 costs 40% less on typical workloads than its predecessor. It also said Fable 5.1’s lower cache-read pricing makes the model around 25% cheaper on typical workloads and up to around 45% cheaper on highly agentic ones. Those workload savings are Anthropic’s estimates, not independent measurements.
The customer-deal figures come from an October 24 report by The Information, as recounted by The Decoder and all-ai.de. DataCamp CEO Jonathan Cornelissen said OpenAI’s offer could lower the company’s costs by 80% to 90% against its current setup. The reporting also described a discount exceeding 30% for Replit on a contract worth more than $100 million a year, and a U.S. bank receiving ChatGPT seats for under $7 each. The available accounts do not establish that these terms are standard or available to other customers.
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 Contract Leverage
For businesses buying AI services, falling list prices and vendor competition can improve their negotiating position. But a model’s per-token rate does not capture a full bill: usage levels, cached tokens, seat counts and contract discounts can change the effective cost. A buyer comparing offers needs to examine the same workload and the same service terms, rather than treating a headline percentage as a universal saving.
The tradeoffs also extend beyond price. Anthropic added capped, expiring API credits to its Max and Team plans, while its Enterprise plan requires a 20-seat minimum. The source account says that requirement puts the service out of reach for many small advisory firms covered by a Schwab arrangement. OpenAI, meanwhile, reduced the usage included with new $200 Pro subscriptions without changing the monthly fee. These changes can make a lower price per token less useful if an organization encounters tighter limits or must buy seats it does not need.
For individual subscribers, the same distinction matters on a smaller scale: a plan’s monthly charge can stay fixed while its included usage changes. List price and included allowance are separate parts of the offer, so readers should check both when comparing services.
Products Worth Considering
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the Price Cuts Built Up
The reported deals follow several public price changes across the summer and early fall. OpenAI reduced the price of GPT-5.6 Luna in July and cut Sol pricing the following month. Anthropic made Sonnet 5’s introductory price permanent in August. In September, OpenAI introduced GPT-6 Sol and Luna at lower rates, Anthropic released Opus 5.5, and OpenAI lowered GPT-6.1 Sol’s cache price. The sequence includes new model launches and promotional changes, not just across-the-board discounts on unchanged products.
Cache pricing can matter particularly for software agents that repeatedly use prior conversation context. Woolf estimates cached tokens account for 96% to 99% of token spending in agentic coding. That is an estimate about those workloads, not a general share for all AI use. It helps explain why a cache-rate reduction can affect an agent’s bill even when a model’s headline input and output rates do not change.
Customers have also described shifting workloads between providers. Rippling CEO Parker Conrad said OpenAI’s share of his company’s coding work rose from 16% in May to 52%, while Anthropic’s share fell from 57% to 18%. Those figures describe one company’s reported usage, not the broader market. Replit CEO Amjad Masad said ChatGPT was better for the company at an October event, then told TechCrunch that Replit remained model agnostic and OpenAI was its preferred partner. TechCrunch also reported that a person familiar with the partnership said it remained in place.
“I think ChatGPT is better now.”
— Replit CEO Amjad Masad, speaking at TechCrunch Disrupt
What Deal Comparisons Still Lack
The full terms of the reported enterprise offers are not public in the source material. It is unclear how long the discounts last, which usage commitments or service conditions apply, and whether the reported rates can be obtained by other customers. The 80% to 90% figure is an executive’s estimate of expected savings against a particular setup, so it cannot be compared directly with a model’s list price.
It is also not established that newer models deliver equivalent results for every customer’s work. OpenAI’s and Anthropic’s claims about capability and workload savings are vendor statements. The source does not provide independent, apples-to-apples tests across the models, nor full data on how changing usage allowances affect customers’ total bills. The reported shifts at Rippling and Replit are individual company developments, not evidence that customers generally are switching providers.
What Buyers Should Track Next
The next useful evidence will be clearer information about the duration and conditions of enterprise discounts, along with updated pricing and plan limits from both vendors. Companies considering a change can compare providers on the same tasks, usage volumes and service requirements, then include token rates, cache charges, seats, credits and caps in the total. Testing matters because a lower-priced successor may not perform equally well on a particular workflow.
Customers should also monitor changes to allowances and renewal terms, not only monthly fees. The available reporting does not say when additional deal details will become public or whether further price changes are planned. For now, the confirmed picture is a fast-moving competition in which published rates are falling, while customized offers and less visible plan rules shape the final cost.
Key Questions
Did OpenAI cut its prices by 90%?
No standard 90% list-price cut is documented here. DataCamp’s CEO said OpenAI’s offer could reduce the company’s overall costs by 80% to 90% compared with its current setup. That is a customer-specific expected saving, not a published discount available to all buyers.
How much did OpenAI’s Sol and Luna prices change?
Based on the pricing figures in the source, 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 $0.10 and $0.50. The prices apply to different model versions.
Did Anthropic lower its model prices too?
Yes. The source reports Opus pricing moving from $5 input and $25 output per million tokens for Opus 5 to $4 and $20 for Opus 5.5. Anthropic also changed cache-read pricing for Fable 5.1 and reported lower estimated costs on certain workloads.
Why can a lower token price fail to lower a customer’s bill by the same amount?
Total cost can depend on usage allowances, cached tokens, seat minimums, expiring credits and negotiated contract terms. A buyer’s savings also depend on whether the newer model performs the same work with similar usage.
Source: ThorstenMeyerAI.com
