OpenAI priced GPT-5 so low, it may spark a price war

OpenAI astounded the tech industry for the second time this week by launching its newest flagship model, GPT-5, just days after releasing two new freely available models under an open source license.

OpenAI CEO Sam Altman went so far as to call GPT-5 “the best model in the world.” That may be pride or hyperbole, as TechCrunch’s Maxwell Zeff reports that GPT-5 only slightly outperforms other leading AI models from Anthropic, Google DeepMind, and xAI on some key benchmarks, and slightly lags on others.

Still, it’s a model that performs well for a wide variety of uses, particularly coding. And, as Altman pointed out, one area where it is undoubtedly competing well is price. “Very happy with the pricing we are able to deliver!” he tweeted.

The top-level GPT-5 API costs $1.25 per 1 million tokens of input, and $10 per 1 million tokens for output (plus $0.125 per 1 million tokens for cached input). This pricing mirrors Google’s Gemini 2.5 Pro basic subscription, which is also popular for coding-related tasks. Google, however, charges more if inputs/outputs cross a heavy threshold of 200,000 prompts, meaning its most consumption-heavy customers end up paying more.

But OpenAI is really undercutting Anthropic’s Claude Opus 4.1, which starts at $15 per 1 million input tokens and $75 per 1 million output tokens. (Anthropic does, however, offer big discounts for prompt caching and batch processing — storing/reusing prompts and processing multiple requests together.)

Anthropic’s model has been extremely popular among programmers, both as a choice within popular coding assistant Cursor and for powering its own such assistant, Claude Code. (Note that Cursor offered GPT-5 as an option minutes after it was announced.)

Developers who have had early access to GPT-5 are touting the pricing. Simon Willison, one of the developers featured in OpenAI’s launch video, writes in his review: “The pricing is aggressively competitive with other providers.”

But GPT-5 is also priced competitively with GPT-4o. OthersideAI’s co-founder and CEO, Matt Shumer (maker of HyperWrite), writes that GPT-5 “is cheaper than GPT-4o, which is fantastic. Intelligence per dollar continues to increase.”

Some on X called OpenAI’s fees for the model “a pricing killer,” while others on Hacker News are offering similar praise.

Will competitors like Anthropic follow? Will Google — who undercut OpenAI on pricing before — get even more affordable? If so, we could be witnessing the start of a much-awaited LLM price war.

There’s no doubt a price war would be welcome. The underlying economics of vibe-coding tool providers, for instance, is pretty shaky because of the high and unpredictable fees they have to pay model makers, as TechCrunch’s Marina Temkin reports. And there are countless startups building on top of AI models as well.

Silicon Valley has been hoping that the LLM price-to-performance ratio will eventually improve, along with inference costs. But it seemed like such an equalization could be years away as the tech industry invests hundreds of billions to build data centers and infrastructure to support growing AI demand. 

OpenAI itself has a $30 billion-per-year contract with Oracle for capacity, when it only recently hit annual recurring revenue of $10 billion. Meanwhile, Meta plans to spend up to $72 billion on AI infrastructure in 2025, and Alphabet has set aside $85 billion for capital expenditures in 2025, driven by AI needs. In the face of such enormous expenses, costs typically go one way: up.

Given such investments, it may be too soon for startups looking at their rising model API bills to rejoice from OpenAI’s lone move to lower pricing. 

Yet this week, OpenAI threw down the gauntlet to put pressure on prices not just once but twice. We’ll see if others follow.