The AI Price War: Comparing OpenAI And Anthropic’s Customer Offers

🔍 Read the full analysis: The AI Price War: Comparing OpenAI And Anthropic’s Customer Offers on ThorstenMeyerAI.com

OpenAI and Anthropic have cut prices on several AI models since June, while OpenAI is also offering discounts to pursue customers using Anthropic, according to reporting by The Information and customer statements. The change matters to businesses budgeting for AI, but reported savings vary by deal and should not be confused with published list prices.

OpenAI’s published prices for its Sol models fell 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, according to the price history compiled by ThorstenMeyerAI.com from vendor announcements and pricing pages. Its Luna model prices moved from $1 and $6 to $0.10 and $0.50 over the same period. Anthropic’s Opus prices shifted from $5 and $25 for Opus 5 in July to $4 and $20 for Opus 5.5, while it also reduced the price of cached input tokens on some models.

Customer reports point to a separate form of price competition: negotiated contracts. The Information reported that OpenAI offered Replit a discount of more than 30% on a contract valued at more than $100 million a year, according to a person cited in its report. DataCamp CEO Jonathan Cornelissen said OpenAI’s offer could reduce his company’s costs by 80% to 90%. That figure describes expected savings against DataCamp’s existing setup, not a general discount from OpenAI’s list price.

Some customers also describe changes in which models they use. 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%. He attributed the shift to cost. Separately, Replit CEO Amjad Masad said at TechCrunch Disrupt on October 16 that the company was no longer working with Anthropic and that he thought ChatGPT was better. He later told TechCrunch that Replit remained model agnostic and OpenAI was its preferred partner; a person familiar with the partnership told TechCrunch it remained in place.

Price Cuts Meet Contract Negotiations

Lower published prices and private discounts can reduce the cost of using AI, particularly for companies running coding agents or other high-volume workloads. But a headline token price is not the whole bill. The cost depends on the model selected, how much input and output it processes, whether cached tokens are used, and what commercial terms apply.

That makes the competition relevant beyond the two vendors. Customers with large contracts may have leverage to seek revised rates or compare offers, while smaller buyers may be more exposed to changes in plan allowances and eligibility rules. A lower price can also lead teams to use models more often, so the final effect on spending depends on actual usage as well as unit cost.

Model substitutions add another variable: a cheaper successor is not automatically equivalent for every task. Buyers need to test the models on their own work and compare the total cost of producing the result they need, rather than assuming that list-price reductions translate directly into matching savings.

How Published Prices Have Shifted

The cited price history, checked against vendor announcements and pricing pages retrieved on October 26, describes a series of changes from June through September. Anthropic introduced Sonnet 5 at an introductory price of $2 per million input tokens and $10 per million output tokens, then made that pricing permanent. OpenAI cut Luna prices in July and later introduced GPT-6 Sol and Luna at prices below earlier levels. In September, Anthropic released Opus 5.5 at $4 and $20 per million tokens.

Some reductions concern cached input, rather than the standard input and output prices. Cached tokens can matter in agentic coding workflows, where a system repeatedly refers to conversation history. Data scientist Max Woolf estimated that cached tokens account for 96% to 99% of token spending in agentic coding; that is his estimate, not a universal measure of AI costs. Anthropic has said its Fable 5.1 model costs around 25% less on typical workloads and up to around 45% less on highly agentic workloads. Those are the company’s estimates, not independently established savings for every customer.

The releases also often replace one model with another rather than reducing the price of an unchanged product. OpenAI has described GPT-6.1 Sol as offering “near-Astra intelligence for a fifth of the price.” Whether a successor performs comparably on a particular company’s tasks requires customer testing.

„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 Deal Terms Do Not Show

The public information does not disclose the full terms of the reported customer agreements, including how long discounts last, which models or usage levels they cover, or whether other contract provisions affect the final bill. The Information’s report is paywalled, and details cited here include its public summary and accounts by The Decoder and all-ai.de.

The broad 80% to 90% figure is especially easy to misread: it is Cornelissen’s estimate of expected savings against DataCamp’s existing costs, not evidence that OpenAI is discounting all customer prices by that amount. The reported Replit discount is more than 30%, but the source account does not establish that every customer can secure similar terms.

Plan limits may also change even when a subscription’s monthly price does not. ThorstenMeyerAI.com reports that OpenAI reduced usage included with new $200 Pro subscriptions and that Anthropic added capped, expiring API credits to Max and Team plans. The supplied material does not detail the size of all allowance changes, their effect on each customer, or whether every new contract will follow the same terms.

What AI Buyers Should Track

Companies considering a renewal or a change of provider can compare written offers, usage limits, credit expiry rules and seat requirements alongside per-token rates. They can also test model performance on representative tasks and track total spending over the same period, so a lower unit price is not mistaken for lower overall costs.

Small organizations should check eligibility requirements before counting on a product or enterprise offer. The source material says Anthropic’s Enterprise plan requires 20 seats, a threshold that may exclude many small advisory firms. It does not provide a complete account of the affected firms or the terms of any alternative plan.

Further pricing-page changes, contract disclosures or customer updates could clarify how widely the reported discounts are available. Until then, the confirmed picture is a run of lower list prices alongside selective negotiated offers and changing plan terms—not a single industry-wide discount or a guaranteed savings rate.

Key Questions

Are OpenAI’s prices down 80% to 90% for all customers?

No. The 80% to 90% figure is DataCamp CEO Jonathan Cornelissen’s estimate of expected savings against his company’s existing setup. It is not a general discount from OpenAI’s published prices.

How did OpenAI’s cited token prices change?

According to the price history cited by ThorstenMeyerAI.com, Sol 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. Luna moved from $1 and $6 to $0.10 and $0.50.

Did Replit leave Anthropic for OpenAI?

Replit CEO Amjad Masad said the company was no longer working with Anthropic and described OpenAI as its preferred partner. He also told TechCrunch that Replit remained model agnostic, and a person familiar with the partnership told TechCrunch it was still in place.

Why do cached-token prices matter?

AI agents may repeatedly refer to prior conversation content, which can involve cached input tokens. Their cost can affect total spending, but the amount varies by workload; the cited estimate that cached tokens make up 96% to 99% of agentic coding spend comes from data scientist Max Woolf.

What should buyers compare beyond list prices?

They should compare actual usage, model performance on their own tasks, cached-token rates, included allowances, credit caps and expiry dates, seat minimums, and the full terms of any negotiated contract.

Source: ThorstenMeyerAI.com