OpenAI Is Poaching Anthropic Customers. Your Model Bill Is Negotiable
OpenAI is winning Anthropic's largest customers with discounts of up to 90 percent, so the list price of a frontier model is now an opening bid rather than a fixed cost. A bootstrapped founder will not get a bespoke contract, but the same fight shows up in public tiers, batch pricing and promotional rates anyone can use.
Key takeaways
- OpenAI reportedly offered Replit more than 30 percent off a contract worth over $100 million a year, according to The Decoder citing The Information.
- DataCamp's CEO said OpenAI's pricing cut his company's costs by 80 to 90 percent.
- Rippling's coding workload moved from 57 percent Anthropic to 18 percent, while OpenAI rose from 16 to 52 percent, its CEO said.
- Public lists already reflect the pressure: gpt-6-luna is $0.10 input and $0.50 output per million tokens, and Batch and Flex tiers are half the standard price.
- A small team cannot negotiate like Replit, so the leverage comes from portability, an evaluation set and pricing that survives a vendor change.
OpenAI is winning Anthropic's largest customers with discounts reported at up to 90 percent, so a frontier model's list price is an opening bid, not a fixed cost (The Decoder). A bootstrapped founder will never see a contract like Replit's, but the same fight appears in public tiers, batch pricing and promotional rates anyone can use. The response is not loyalty to a lab. It is making the model underneath replaceable and pricing so a vendor change does not wipe out margin.
TL;DR
- OpenAI reportedly offered Replit more than 30 percent off a contract worth over $100 million a year (The Decoder).
- DataCamp's CEO said OpenAI's pricing cut costs by 80 to 90 percent (The Decoder).
- Rippling's coding workload moved from 57 percent Anthropic to 18 percent, while OpenAI rose from 16 to 52 percent (The Decoder).
- Public lists show the pressure: gpt-6-luna is $0.10 input and $0.50 output per million tokens, and Batch and Flex tiers are half price (OpenAI pricing).
Disclosure: ZeroRaised publishes this analysis and is a launch site for bootstrapped products. Facts come from the linked sources as checked on 26 Oct 2026; ZeroRaised has no commercial relationship with the companies named.
Model portability is the ability to move a workload from one model provider to another by changing configuration and prompts rather than rewriting the product. It is the main defence a small team has when the price of a model is set by a negotiation it cannot join.
What did the reported deals show?#
The Decoder, citing The Information, reported that OpenAI is pulling major customers from Anthropic with aggressive discounts (The Decoder). Rippling CEO Parker Conrad said OpenAI's share of its coding workload rose from 16 to 52 percent while Anthropic's fell from 57 to 18 percent (The Decoder). Replit switched after OpenAI offered more than 30 percent off a contract worth over $100 million a year (The Decoder). DataCamp CEO Jonathan Cornelissen said OpenAI's pricing cut his costs by 80 to 90 percent (The Decoder).
The same pattern appears in seats. A US bank with more than $7 billion in assets got ChatGPT seats doubled to 500 employees at no extra charge, paying under $7 per user instead of the usual $20 or more (The Decoder). Treat the percentages as the shape of the market, not a price list.
The public prices already moved#
You do not need an enterprise contract to benefit. The public table is where the pressure becomes usable.
| Model, standard tier, short context (per million tokens) | Input | Output | Source |
|---|---|---|---|
| gpt-6-astra | $10.00 | $50.00 | OpenAI pricing |
| gpt-6.1-sol | $2.00 | $10.00 | OpenAI pricing |
| gpt-6-luna | $0.10 | $0.50 | OpenAI pricing |
| gpt-5.6-sol | $4.00 | $20.00 | OpenAI pricing |
Cached input on gpt-6.1-sol is $0.10 per million tokens, a tenth of the standard input rate, and Batch and Flex are half price, so it drops to $1.00 input and $5.00 output (OpenAI pricing).
Long context above 272K input costs more, with gpt-6.1-sol at $4.00 input and $15.00 output, and regional endpoints carry a 10 percent uplift for models released on or after 5 March 2026 (OpenAI pricing). The GPT-5.6 Sol promotional price is only promised "at least through November 21, 2026" (OpenAI pricing).
Why the labs can discount now#
Three things push prices down at once. OpenAI made real cuts through its Sol and Luna model lines, not only through negotiated deals (The Decoder). Anthropic's growth slowed from the second quarter to the third, according to an investor memo cited in the report (The Decoder). And both face pressure from Chinese models such as DeepSeek and Kimi (The Decoder).
Kimi K3.1 Pro lists at $2.00 input and $10.00 output per million tokens with a 1,048,576-token context (Kimi pricing). Our analysis of Kimi K3.1 Pro and open weights covers the self-hosting angle. Competition is working for buyers.
The discounts a small team can actually use#
You will not get Replit's contract. You can still take the same levers in public form.
- Use Batch and Flex for work that can wait. Both are half the standard price (OpenAI pricing). Nightly summaries, backfills and evaluations belong there.
- Cache the repeated part of your prompt. Cached input on gpt-6.1-sol is $0.10 instead of $2.00 (OpenAI pricing). With a stable system prompt, that is the largest single saving in most chat products.
- Right-size the model. gpt-6-luna at $0.10 input and $0.50 output is a twentieth of the mid-tier input price (OpenAI pricing). Classification and routing rarely need the top model.
- Watch the promotional calendar. Note the end date next to any price you build on and work out margin at the later price too (OpenAI pricing).
- Use startup credits while they last. Credits buy time, not a business model; our guide to Anthropic credits for startups covers one programme and its limits.
The mechanics of each lever are in how to reduce LLM costs.
Do not price your product on someone else's discount#
The tempting mistake is passing a promotional model price through as a permanent low price. If the discount ends, margin disappears and you must raise prices on people who joined for the cheap one.
Set retail price on a model cost you would still accept after a 50 percent increase, and keep the difference for support, failed payments and unbuilt features. If that makes you uncompetitive, narrow the scope instead of chasing a cheaper vendor.
Unlimited plans are the other trap. Your heaviest users set your costs, and a vendor rise turns your best customers into your worst margin. Usage limits or credits keep cost and revenue linked. Our SaaS pricing guide covers plan shape, and how to launch an AI app covers testing before strangers use it.
What this means for bootstrapped founders#
- Put every call behind one interface. One client that takes a model name, not provider SDK calls scattered through the code.
- Build a five task eval. Real requests, known good answers, cost per task. That is your negotiating data even when there is no negotiation.
- Keep a second provider warm. Send a small share of traffic to it, so an outage is not the moment you discover the integration broke.
- Price with headroom. Assume today's rate is a discount, and check that margin survives a doubling.
- Read renewal terms at your scale. Monthly commitments, credit expiry and auto-reload limits matter more than a headline percentage. Our guide to rotating API keys keeps the account tidy.
None of this requires volume, only treating the model as a supplier you can change.
Where ZeroRaised fits#
ZeroRaised is a launch site for products built without institutional VC or accelerator equity. If you have a product whose costs and pricing you understand, you can submit it here and get a page you can point people to. There is no promise of traffic or results. For the next analysis, join the ZeroRaised newsletter.
Sources#
Frequently asked questions
What is the OpenAI and Anthropic discount war?
It is a reported price competition between the two labs for large enterprise accounts. The Decoder, citing The Information, reported that OpenAI is offering aggressive discounts to pull major customers from Anthropic, including more than 30 percent off a contract worth over $100 million a year for Replit.
Can a bootstrapped founder get the same discounts?
Not the bespoke enterprise deals, which depend on contract size and volume. A small team can still use the public versions of that pressure: cheaper model tiers, cached input, Batch and Flex pricing at half the standard rate, startup credits and promotional rates with published end dates.
Should I switch my product to the cheapest model?
Only after your own tests. Price per token is not cost per task, because a model that uses more tokens or more calls can cost more overall. Build a small evaluation set from real requests, measure cost per task, and shop on that number rather than the price list.
How do I protect against a price change?
Keep model calls behind one interface, keep prompts in your codebase, and maintain a tested second provider. Then a price rise is a configuration change rather than a rewrite. Also avoid selling a flat unlimited plan whose margin depends on one vendor's current discount.