🔍 Read the full analysis: What Customers Should Know About OpenAI And Anthropic’s Price Cuts on ThorstenMeyerAI.com
Get the latest gadgets 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 prices for several AI models since June 2026, while OpenAI is also offering discounts to some large customers, according to reporting cited by the source. Headline savings do not apply uniformly: some figures describe negotiated deals or estimated total cost reductions, and subscription allowances and contract terms are changing too.
OpenAI and Anthropic have reduced prices for several AI models since June, and OpenAI is offering some large customers additional discounts, according to the vendor pricing pages and reporting cited by ThorstenMeyerAI.com on October 26. The changes may lower bills for businesses and developers, but reported savings vary by customer and usage: a widely cited 80% to 90% cost reduction is one CEO’s estimate of total savings, not a general discount on OpenAI’s list prices.
OpenAI’s listed token prices changed between June and September. According to the source’s comparison of vendor announcements and pricing pages retrieved October 26, the Sol model line 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. These are prices for different model versions, rather than a like-for-like reduction on an unchanged model.
Anthropic also changed its pricing. The source reports that Opus 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. It also says Anthropic cut cache-read pricing for some models. Anthropic’s claims that Opus 5.5 costs 40% less on typical workloads and that Fable 5.1 can cost less on agentic workloads are estimates from the company, not independently established savings for every customer.
Separately, The Information reported on October 24 that OpenAI was using discounts to attract customers from Anthropic. The source cites a reported offer of more than 30% off for Replit on a contract valued at over $100 million a year, and a US bank receiving sub-$7 seats. Those are reported, customer-specific terms, not published rates available to all buyers. The source also reports that OpenAI’s new $200 Pro subscriptions include less usage than before, while Anthropic added capped, expiring API credits to its Max and Team plans and set a 20-seat minimum for Enterprise.
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 Complex Bills
Lower token prices can reduce costs for companies running chatbots, coding tools and other AI workloads. But a price per million tokens is only one part of the bill. Model choice, cached-token pricing, usage allowances, credits and negotiated rates can change what an organization pays in practice. A cheaper model may also perform differently on a particular task, so buyers need to evaluate both cost and results on their own workloads.
The terms matter to smaller organizations as well as large buyers. The source says Anthropic’s 20-seat Enterprise minimum could rule out many small advisory firms, while capped or expiring credits can limit the value of a plan even if its monthly fee is unchanged. OpenAI’s reported reduction in included usage for new $200 Pro subscriptions illustrates the same point for individual subscribers: a stable monthly price does not necessarily mean stable access.
For procurement teams, the reported negotiations suggest that asking vendors for revised terms may be worthwhile, especially when there are competing options. That does not mean every buyer can obtain the discounts reported for major customers. Businesses should compare the full contract and measure costs on representative work, rather than treating a single headline rate as a reliable forecast of their savings.
Products Worth Considering

Samsung Odyssey G55C 32-Inch QHD Curved Gaming Monitor

LG UltraGear 34-Inch Curved QHD Gaming Monitor, 160Hz, FreeSync Premium

Samsung 27-Inch Odyssey G5 QHD Gaming Monitor (G51F)
As an affiliate, we earn on qualifying purchases.
How Prices Changed Since June
The source describes a sequence of price and product changes from June to September 2026. Anthropic launched Sonnet 5 in June at introductory pricing of $2 per million input tokens and $10 per million output tokens, then made that pricing permanent in August, according to the source. OpenAI reduced Luna pricing in July and lowered Sol pricing for a limited period in August before September releases brought further changes. Anthropic launched Opus 5 in July and Opus 5.5 in September at lower listed rates.
Some reductions involve cache reads, which can matter for AI agents that repeatedly use earlier conversation content. The source cites data scientist Max Woolf’s estimate that cached tokens can make up 96% to 99% of token spending in agentic coding. That is an estimate from Woolf, not a universal figure for every application. Anthropic has said its cache pricing and model changes reduce costs on certain workloads; customers should test whether those assumptions fit their own usage.
Customer movement is also part of the reported picture. 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%. He attributed the shift to cost, according to the source. Replit CEO Amjad Masad said publicly that the company was no longer working with Anthropic and that he thought ChatGPT was better, then told TechCrunch that Replit remained model-agnostic and OpenAI was its preferred partner. TechCrunch also cited a person familiar with the partnership who said it remained in place.
Which Savings Apply to Buyers
Most negotiated contract terms are not public, and the reported discounts do not establish what another customer can obtain. The source notes that the 80% to 90% figure attributed to DataCamp refers to expected overall costs, while the precise comparison and workload details are not fully laid out in the available account. The reported Replit discount and bank seat price are also specific deals.
It is not yet clear how broadly OpenAI’s customer-specific offers are available, how long they last, or what usage commitments and other contract conditions apply. Nor do lower rates for successor models show that each one will deliver comparable performance on every customer’s tasks. Subscription usage limits and credit rules can also change, so buyers need current contract details rather than relying on older plan descriptions.
What Buyers Should Check Next
Businesses reviewing AI contracts should ask vendors for current pricing, usage limits, credit expiration rules, minimum seat counts and any requirements attached to discounts. They can compare providers using representative workloads and include input, output and cached-token charges in the calculation. Where feasible, keeping work portable across more than one provider can give teams flexibility if prices or product performance change.
Individual subscribers should check the allowance included in their current plan and whether it differs from the terms for new subscribers. The source provides no announced date for a further round of price changes or a broad publication of the reported negotiated deals. The next useful indicators will be updated vendor pricing and plan pages, clearer contract terms, and evidence of how the newer models perform on customers’ actual work.
Key Questions
Are OpenAI’s prices down by 80% to 90% for everyone?
No. The source describes that range as DataCamp CEO Jonathan Cornelissen’s estimate of total savings against his existing setup. It is not a general 80% to 90% reduction in OpenAI’s public prices.
Which OpenAI model prices changed?
According to the source’s comparison of vendor pricing pages, 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. The comparisons involve different model versions.
Did Anthropic lower prices too?
Yes. The source reports that Opus pricing changed from $5/$25 per million input/output tokens for Opus 5 to $4/$20 for Opus 5.5. It also reports changes to cache-read pricing and credits. Check Anthropic’s current pricing and plan terms before estimating a bill.
Do lower list prices guarantee a lower monthly bill?
No. A customer’s total depends on usage, model choice, cached tokens, subscription allowances, credits and negotiated contract terms. A newer, lower-priced model may also perform differently on a specific task.
What should customers compare before switching?
Compare the full cost on representative work, including input, output and cache charges, along with usage caps, credit expiration, seat minimums and contract conditions. Confirm that the model meets the required quality and performance needs.
Source: ThorstenMeyerAI.com
