🔍 Read the full analysis: Why AI Customers Are In The Middle Of OpenAI And Anthropic’s Price Fight on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI is using customer-specific discounts to compete for enterprise AI business as it and Anthropic reduce model prices. The reported savings vary: The Information described a discount above 30% for Replit, while DataCamp’s CEO estimated 80% to 90% lower total costs under an OpenAI offer. Usage limits, credits, seat minimums and model performance can change the real cost.
OpenAI is offering discounts to draw enterprise customers away from Anthropic, according to an October 24 report by The Information, putting companies that buy AI models at the center of a price fight. The reported offers range from a discount above 30% for Replit to an estimated 80% to 90% reduction in total costs for DataCamp, but that higher figure is not a discount off OpenAI’s listed prices.
The companies have also cut published model prices in recent months. A review of vendor pricing pages cited by ThorstenMeyerAI.com puts OpenAI’s Sol model at $2 per million input tokens and $10 per million output tokens in September, down from $5 and $30 in June. Its smaller Luna model fell from $1 and $6 to 10 cents and 50 cents per million tokens. These comparisons involve successive models and price changes, not necessarily the same model becoming cheaper.
Anthropic also reduced prices. Its Opus line moved from $5 per million input tokens and $25 per million output tokens for Opus 5 in July to $4 and $20 for Opus 5.5. Anthropic also lowered the cost of reading cached tokens, which can affect the bill for systems that repeatedly process long conversations. The company says its newer models can cost less on typical workloads, but those estimates are its own and may not apply to every customer.
The reported customer changes suggest price is affecting vendor selection. Rippling CEO Parker Conrad said OpenAI’s share of the company’s coding work rose from 16% in May to 52%, while Anthropic’s share fell from 57% to 18%. The Information also reported that OpenAI offered Replit a discount of more than 30% on a contract valued at over $100 million a year. These are customer-specific examples, not evidence that every buyer has received similar terms.
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
How Discounts Could Reshape AI Buying
For companies using AI across engineering and other business functions, vendor pricing can affect both budgets and which models employees use. A lower offer may let a buyer expand access or shift workloads, as the reported Rippling figures indicate. But the headline discount alone does not establish the total value of a contract.
Usage allowances, seat counts and credit rules can change the effective price. ThorstenMeyerAI.com reports that Anthropic added capped, expiring API credits to some Max and Team plans and that its Enterprise plan requires at least 20 seats. The same report says OpenAI reduced the usage included with new $200 Pro subscriptions without changing the subscription price. These terms matter to smaller organizations and individual users as well as large firms.
The competition gives buyers a reason to compare offers and test alternatives, but switching carries operational questions: whether a model performs well on a company’s work, how usage is measured, and whether contract terms preserve access at expected volumes. A low per-token rate may not translate into lower costs if a workload consumes more tokens or requires a different level of service.
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)
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
As an affiliate, we earn on qualifying purchases.
A Rapid Run of Model Price Changes
The price moves followed a series of new model releases and revisions from June through September. The pricing timeline assembled by ThorstenMeyerAI.com draws on vendor announcements and pricing pages checked on October 26, as well as pricing analysis by data scientist Max Woolf. It shows several changes across both companies, including promotional prices that later became permanent and lower prices for newer models.
Model prices are only part of the bill for automated coding systems. These systems may repeatedly send earlier conversation content back to a model. Woolf estimates cached tokens account for 96% to 99% of token spending in agentic coding workloads. That estimate explains why cache pricing can matter, though the share will depend on the workload. Anthropic says its Fable 5.1 model costs about 25% less on typical workloads and up to about 45% less on highly agentic ones; those are company estimates, not independent measurements.
The October reporting adds a new dimension to the public price cuts: private deals aimed at specific customers. The Information’s article, “Anthropic’s Margin is OpenAI’s Opportunity,” was published October 24. Its public summary and coverage by The Decoder and all-ai.de describe the offers and customer comments. The article itself is paywalled, and the details available publicly do not provide a full account of each contract.
“OpenAI’s share of the company’s coding work rose from 16% in May to 52%, while Anthropic’s fell from 57% to 18%.”
— Rippling CEO Parker Conrad, as reported by The Information
What the Deal Figures Do Not Show
The full terms of the reported customer agreements are not public in the source material. It is unclear how long the discounts last, which usage volumes or services qualify, and whether they include commitments or other conditions. The reported 80% to 90% figure for DataCamp is a CEO’s estimate of total savings against its current setup; the available reporting does not detail the calculation.
There is also no basis here for treating the reported offers as standard prices available to all customers. Nor do the examples establish how broadly companies are switching vendors. Replit’s public comments include both a preference for OpenAI and a statement that it remains model agnostic, while TechCrunch reported that a person familiar with the partnership said it was still in place.
Published prices and customer reports also do not settle how models compare on accuracy, reliability or cost for a particular task. Newer models replace older ones in several of the price comparisons, so the lower rate does not by itself show that equivalent work can be completed at the same quality or token volume.
What Buyers Should Track Next
The next evidence will come from more customer disclosures, updated vendor prices and the terms attached to renewed or expanded contracts. Businesses considering a switch can compare total bills across representative workloads, including input, output and cached-token use, rather than relying on a single headline rate. They can also check seat minimums, included usage, credit expiration and the duration of promotional pricing.
For current customers, the reported offers may provide a basis to ask vendors for revised terms, but the available reporting does not show how either company will respond across the wider market. Buyers will also need to test whether a proposed replacement meets their performance needs before moving important work. The balance between public prices and private discounts is likely to remain central as OpenAI and Anthropic compete for business.
Key Questions
Is OpenAI offering every customer a 90% discount?
No. The reported 80% to 90% figure is DataCamp CEO Jonathan Cornelissen’s estimate of total savings against the company’s existing setup. The source material does not describe it as a discount from OpenAI’s list price or as a standard offer for all customers.
What discount did OpenAI reportedly offer Replit?
The Information reported an offer of more than 30% on a contract valued at over $100 million a year, citing a person familiar with the matter. The public information does not provide the full contract terms.
Have companies moved coding work from Anthropic to OpenAI?
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%. The figures describe Rippling, not the market as a whole.
Why do usage limits matter if subscription prices stay the same?
A plan’s effective value depends on how much usage it includes and what happens when limits are reached. ThorstenMeyerAI.com reports that OpenAI cut the usage included in new $200 Pro subscriptions without changing the price, illustrating why buyers should check allowances as well as fees.
Do lower token prices guarantee a lower AI bill?
No. Costs depend on workload, token volume, cached-token use, model performance and contract terms. Several cited price comparisons involve new models replacing older ones, so customers need to test costs and results on their own tasks.
Source: ThorstenMeyerAI.com
