Skip to content
NewsFunding & M&A

OpenAI discounts of up to 90 percent pull named Anthropic customers toward GPT

· by Pondero Newsdesk

The short version

Rippling, Replit, and DataCamp are named customers shifting coding workload from Anthropic to OpenAI after discounts as steep as 90 percent, per an October 25, 2026 report relayed by The Decoder.

OpenAI discounts of up to 90 percent pull named Anthropic customers toward GPT

Rippling's coding-model mix flipped within months: OpenAI's share of the company's workload rose from 16 percent to 52 percent while Anthropic's fell from 57 percent to 18 percent, according to Rippling CEO Parker Conrad, cited in an October 25, 2026 report from The Decoder that relays reporting from The Information.

What

OpenAI is cutting enterprise API prices by as much as 90 percent specifically to win customers away from Anthropic, and named accounts are moving. Replit switched model suppliers after OpenAI offered more than 30 percent off a contract worth over $100 million a year, per The Decoder. DataCamp CEO Jonathan Cornelissen said OpenAI's pricing cut his company's AI costs by 80 to 90 percent. A US bank with more than $7 billion in assets got its ChatGPT seat count doubled to 500 employees at no extra charge, bringing its per-user cost under $7 versus the usual $20 or more, according to the same report.

OpenAI is partly buying market share at the expense of its own margins, but the report also credits real efficiency gains: price cuts running through what it calls its Sol and Luna model lines, built on more efficient models and infrastructure rather than discounting alone. (GPT 5.6-Sol, part of that line, shows up as a named comparison point in a separate DeepSeek benchmark release discussed below, confirming Sol as an active production model rather than a placeholder name.) Anthropic's own growth slowed from the second quarter to the third, according to an investor memo cited in the report, and both companies face added price pressure from China, where DeepSeek and Kimi's newer models deliver comparable performance for a fraction of the cost.

Enterprise buyers are now switching primary suppliers on price alone

The Rippling, Replit, and DataCamp examples share one trait: none of the named executives cited a capability gap as the reason for moving spend. The swap was priced, not benchmarked. That is a different competitive dynamic than the one AI labs have mostly competed on since 2023, where buyers picked a model for eval scores and stuck with it through multiple price increases. If a company with an eight-figure or nine-figure AI contract will reallocate 30 to 40 percentage points of workload (the Rippling swing cited above) for a 30 to 90 percent discount, procurement teams at other large accounts have reason to ask their own vendors for the same terms, or to start splitting workload across two providers to keep more negotiating power going into the next renewal.

The timing adds pressure for Anthropic specifically. The company is heading toward a planned November IPO, and its third-quarter revenue of $16.8 billion marked roughly a 45 percent jump from the second quarter, a sharp deceleration from the more than 140 percent jump it posted from the first quarter to the second. A public discounting war with OpenAI, playing out through named customers rather than press releases, is not the growth story a company typically wants in front of IPO investors weighing its pricing power.

Context and reactions

Neither OpenAI nor Anthropic issued a statement addressing the specific discount terms described in the report; the figures and quotes come from the named customers and the original reporting, not from either vendor's own communications. The pattern follows a broader 2026 trend of frontier labs cutting prices as inference costs fall and Chinese open-weight models narrow the capability gap at much lower cost. On the DeepSWE coding-agent benchmark, DeepSeek's V4.1-Flash-1017 release scored 77.0 percent at an average cost of under $0.20 per task, while GPT-6 Astra, Gemini 3.8 Flash, and Claude Opus 5 each ran between $2 and over $10 per task in the same comparison, according to DeepSeek's own published benchmarks. That gap illustrates why Anthropic and OpenAI are both under pressure on price from two directions at once: each other, and the open-weight Chinese labs undercutting both on a per-task basis. Anthropic answered one round of that pressure in September with Opus 5.5, which The Decoder has reported matches rival Fable 5.1 at a 40 percent lower cost, but that model-side response does not touch the enterprise contract-pricing fight described in this report.

What to watch next

Whether Anthropic responds with its own enterprise discounting, rather than relying on per-model price cuts, and whether more named customers beyond Rippling, Replit, and DataCamp confirm reallocating workload specifically over contract terms. Anthropic's IPO filings, once public, should also show how much revenue concentration sits in accounts now exposed to this kind of competitive discounting. If DeepSeek, Kimi, or another open-weight lab keeps closing the per-task cost gap shown on benchmarks like DeepSWE, both US labs may face pressure to extend similar discounts well beyond the handful of named accounts in this report.

Sources