🔍 Read the full analysis: OpenAI Or Anthropic? Compare The Price Cuts Before You Sign on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI and Anthropic cut several model prices in 2026, while reports describe OpenAI offering discounts to win business from Anthropic. Reported customer savings vary: the widely cited 80% to 90% figure is one CEO’s expected reduction in total costs, not a general list-price discount. Buyers still need to compare model performance, usage rules, credits, seats and contract terms.
OpenAI and Anthropic have reduced prices on several AI models since June, and OpenAI is also offering customer-specific discounts as it seeks business from Anthropic, according to a report by The Information. The changes give companies more room to negotiate, but the headline price alone does not show the full cost: usage limits, credits, seat minimums and model performance can all affect what a buyer pays.
Vendor pricing pages reviewed by ThorstenMeyerAI.com on October 26 show OpenAI’s Sol model line falling 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 line moved from $1 and $6 per million tokens to $0.10 and $0.50. These are prices for successive models, not simply a reduction in the price of an unchanged model.
Anthropic also lowered some prices. 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 reduced cache-read prices on some offerings. The company’s claims that Opus 5.5 costs 40% less on typical workloads and that its Fable 5.1 model costs less on certain agentic workloads are vendor estimates, not independent measurements.
The customer offers reported by The Information differ from the public list prices. Replit reportedly received a discount of more than 30% on a contract worth over $100 million a year. DataCamp CEO Jonathan Cornelissen said OpenAI’s offer could reduce his company’s total costs by 80% to 90%; that figure describes expected savings against its existing setup, not an 80% to 90% reduction from OpenAI’s published rates. A U.S. bank was reportedly offered seats for less than $7 each, though the available source material does not specify all contract 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
Why AI Buyers Have More Leverage
Lower list prices and vendor competition can reduce the cost of running AI services, particularly for companies with large token volumes or substantial enterprise contracts. The reported offers also suggest that large buyers may be able to negotiate terms beyond public pricing. That can make it worthwhile for procurement teams to request competing bids and revisit existing agreements rather than assume posted rates are final.
But a lower per-token figure does not guarantee a lower bill. In agentic coding, systems may repeatedly process prior conversation context; the source article cites an estimate that cached tokens account for 96% to 99% of token spending in such work. Cache pricing and how a vendor counts usage can therefore matter as much as headline input and output rates. Companies should compare costs on their own workloads, while checking whether a cheaper successor model performs well enough for the tasks it would handle.
Subscription buyers face a similar issue. The source reports that OpenAI reduced the usage included with new $200 Pro subscriptions without changing the monthly price, while Anthropic introduced capped, expiring API credits for some plans. A stable subscription fee may therefore buy a different amount of service over time.
Products Worth Considering
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Price Cuts and Contract Changes
The price reductions accumulated over several months rather than arriving as a single announcement. Max Woolf, a data scientist who tracks language-model pricing, outlined the sequence in an October 25 blog post. ThorstenMeyerAI.com says it checked the figures against vendor announcements and pricing pages retrieved on October 26.
Some reductions were attached to new or successor models; others concerned promotional pricing or cached inputs. Anthropic made Sonnet 5’s introductory price permanent, while OpenAI cut Luna prices and later introduced GPT-6 models at lower rates than earlier offerings. The distinction matters: model capability can change alongside price, so comparing two price points without testing the models may not reveal equivalent value.
Reported customer movements indicate that the competition is affecting deployment choices, though they do not establish a broad market shift. Rippling CEO Parker Conrad said OpenAI’s share of his company’s coding work rose from 16% in May to 52%, while Anthropic’s fell from 57% to 18%. Replit CEO Amjad Masad described OpenAI as its preferred partner, while Replit later told TechCrunch that it remained model agnostic. Those statements describe particular companies, not all enterprise customers.
“I think ChatGPT is better now.”
— Amjad Masad, Replit CEO, speaking at TechCrunch Disrupt
Terms Behind the Reported Offers
The public reporting does not provide complete contract documents for the customer-specific offers, so it is unclear how long the discounts last, what usage commitments apply or whether additional fees are attached. The Information’s article is paywalled; the source article says its account relied on the public summary and detailed coverage by The Decoder and all-ai.de. The 80% to 90% estimate should not be read as a generally available OpenAI discount.
It is also unclear whether the reported shifts in model use will persist or reflect a broader trend. Companies may choose different models for different tasks, and reported workload shares do not by themselves establish which service is cheaper or more capable across an entire business. Anthropic’s performance and workload-cost figures are company claims; buyers would need to test them against their own usage.
Plan details may change. The source reports that Anthropic’s Enterprise plan requires 20 seats, potentially limiting access for smaller firms, and that some Max and Team plans now include capped, expiring API credits. It also says a U.S. bank’s offer involved 500 seats at no extra charge, but the account is truncated in the supplied source material and the full conditions are not available.
What Buyers Should Check Next
Companies considering a new contract should request written, comparable proposals from both providers and test candidate models on representative workloads. A useful comparison should include token usage, cache reads, output quality, latency and any costs associated with retries or additional tools—not just the published price per million tokens.
Procurement teams should also confirm how long discounts remain in force, whether they require minimum spending or seat commitments, how usage is capped, and whether unused credits expire. For existing customers, the latest price and plan terms should be checked against the current contract before assuming a public rate applies. The next meaningful evidence will be clearer contract details and customer results showing whether the savings hold after usage and performance are measured.
Key Questions
Did OpenAI cut its prices by 90%?
No general 90% list-price cut is established by the source material. DataCamp CEO Jonathan Cornelissen described an expected 80% to 90% reduction in total costs against the company’s existing setup; that is different from a discount of that size on OpenAI’s published prices.
How much did OpenAI’s listed token prices fall?
For the model generations compared 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 comparisons involve successive models, not necessarily identical systems.
Did Anthropic also lower prices?
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 reduced some cache-read prices and made Sonnet 5’s introductory pricing permanent.
What should businesses compare before signing?
Compare tested performance and total workload cost, including input and output tokens, cached reads, usage allowances, credits, seat minimums, contract duration and discount conditions. A low list price may not produce the lowest bill if the plan has restrictive limits or the model needs more usage to complete the same work.
Source: ThorstenMeyerAI.com
