The AI Price War Is Heating Up Between OpenAI And Anthropic

🔍 Read the full analysis: The AI Price War Is Heating Up Between OpenAI And Anthropic on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

OpenAI and Anthropic have cut prices for several AI models since June, and OpenAI is also offering reported discounts to some large customers as competition for business accounts intensifies. The changes could lower costs for companies using AI, but reported savings vary by customer and do not necessarily reflect reductions in public list prices, a dynamic also reflected in the frontier model price war.

According to pricing information and announcements cited by ThorstenMeyerAI.com, OpenAI’s Sol model line moved from $5 input and $30 output per million tokens in June to $2 and $10 for GPT-6.1 Sol in September. Its Luna line fell from $1 and $6 per million tokens to $0.10 and $0.50. These are prices for successive models, rather than proof that the same model became cheaper while remaining unchanged.

Anthropic also reduced prices across model releases. The source reports that Opus went from $5 input and $25 output per million tokens for Opus 5 in July to $4 and $20 for Opus 5.5. It also cites lower cache-read pricing on newer models. Anthropic has said those changes can reduce costs on typical and heavily agentic workloads; those savings estimates are the company’s own claims, not independently established results for every customer.

Separately, The Information reported on October 24 that OpenAI is using discounts to compete for large customers, as explored in this account of OpenAI and Anthropic’s enterprise push. The reported examples include a discount of more than 30 percent on a Replit contract valued at over $100 million a year, and a U.S. bank receiving 500 ChatGPT seats for less than $7 per seat. The account of the bank’s deal and its terms comes from the source article’s reporting, not a public contract.

Lower Prices Reshape Enterprise AI Buying

For companies, lower per-token rates and negotiated discounts can reduce the cost of coding tools, customer services and other AI workloads. The shift also gives buyers a reason to revisit contracts: a price quoted months ago may no longer reflect what a vendor will offer to retain or win an account. The reported movement in customers’ usage suggests price and model performance are both shaping purchasing decisions.

List prices alone do not determine a bill. Some AI products charge by usage, while business plans may bundle seats, credits or allowances. The source says Anthropic added capped, expiring API credits to its Max and Team plans and requires 20 seats for Enterprise. It also reports that OpenAI reduced usage included with new $200 Pro subscriptions without changing the subscription price. These terms can affect effective costs even when a monthly fee or headline token rate appears stable.

That makes the current competition relevant to both procurement teams and individual subscribers. Organizations can compare vendors and test workloads against their own requirements, rather than assuming a lower list price means a lower total bill. Users should also check included limits, since plan allowances can change separately from subscription prices.

How the Price Cuts Accumulated

The reductions described in the source came through multiple launches and price changes over several months, rather than one announcement. OpenAI cut Luna pricing in July, reduced Sol pricing in August, and introduced lower-priced GPT-6 models in September. Anthropic launched Sonnet 5 at an introductory rate in June and made that pricing permanent in August, then released Opus 5.5 in September at lower listed rates than Opus 5.

Data scientist Max Woolf published a timeline of model pricing on October 25. ThorstenMeyerAI.com says it checked the listed figures against vendor announcements and pricing pages accessed October 26. A further pricing factor is prompt caching: coding agents may repeatedly process prior conversation content, so cache rates can matter alongside standard input and output rates. Workloads differ, however, and a vendor’s estimate of savings does not guarantee the same result for each buyer.

The customer examples also need careful reading. Rippling CEO Parker Conrad said his company’s share of coding work handled by OpenAI models rose from 16 percent in May to 52 percent, while Anthropic’s share fell from 57 percent to 18 percent. He cited cost as a reason. Replit CEO Amjad Masad said at an October 16 event that the company was no longer working with Anthropic and described ChatGPT as better; he later told TechCrunch that Replit remained model agnostic and OpenAI was its preferred partner. TechCrunch separately reported that a person familiar with the partnership said it was still in place.

“I think ChatGPT is better now.”

— Amjad Masad, Replit CEO, speaking at TechCrunch Disrupt on October 16

Deal Terms and Savings Vary

The full terms of several customer agreements are not public in the material provided. The reported 80-to-90-percent savings for DataCamp is an executive’s expected reduction in total costs, not evidence that OpenAI cut its list prices by that amount. The basis for comparing those costs, including the workloads and services involved, is not fully detailed.

It is also unclear how broadly the reported discounts are available, how long they last, or what usage commitments customers must make. The source does not establish whether other buyers can obtain comparable terms. For Replit, public comments about its relationship with Anthropic differ in emphasis, and TechCrunch reported that the partnership remained in place despite Masad’s remarks.

Finally, lower prices across newer models do not by themselves establish equivalent performance for every task. The source notes that buyers need to test successor models on their own workloads. Vendor claims about typical savings and model capability should be distinguished from a customer’s measured results.

What Buyers Should Track Next

Companies are likely to keep comparing model quality, usage costs and contract terms as vendors compete for enterprise accounts. Buyers considering a renewal or new agreement can ask for current pricing, clarify discount duration and commitments, and compare projected bills against their actual workloads. Testing more than one provider can help show whether a cheaper model performs adequately for a specific task.

Customers should also watch the terms surrounding headline prices: seat minimums, token allowances, cache rates, credit caps and expiration dates. For individual subscribers, the published monthly fee may not reveal how much usage is included. Further price changes or newly disclosed customer agreements could clarify whether current discounts are isolated deals or part of a broader sales strategy; the source material does not establish that yet.

Key Questions

Have OpenAI and Anthropic both lowered their AI model prices?

Yes. The source describes multiple price reductions and lower-priced model releases from both companies between June and September 2026. The figures compare different model versions in several cases.

Did OpenAI offer customers a 90 percent list-price discount?

The source does not establish that. DataCamp CEO Jonathan Cornelissen described expected total cost savings of 80 to 90 percent against his company’s current setup. That is not the same as a 90 percent reduction in OpenAI’s public list price.

Why can cache pricing affect AI costs?

Some workloads repeatedly process earlier conversation content. The source says cached tokens can make up much of the token spend in agentic coding, so cache-read rates may affect the total bill. The effect depends on the customer’s workload.

Are OpenAI and Anthropic discounts available to every customer?

The available reporting does not say that. Some deals appear to be individually negotiated, and their full terms are not public. Buyers should request current terms directly from vendors.

What should companies compare besides list prices?

They should compare expected costs for their own workloads, model performance, seat requirements, included usage, credit limits and expiration rules. A lower per-token price does not automatically mean a lower overall bill.

Source: ThorstenMeyerAI.com