OpenAI astounded the tech business for the second time this week by launching its latest flagship mannequin, GPT-5, simply days after releasing two new freely out there fashions underneath an open supply license.
OpenAI CEO Sam Altman went as far as to name GPT-5 “one of the best mannequin on this planet.” Which may be satisfaction or hyperbole, as TechCrunch’s Maxwell Zeff stories that GPT-5 solely barely outperforms different main AI fashions from Anthropic, Google DeepMind, and xAI on some key benchmarks, and barely lags on others.
Nonetheless, it’s a mannequin that performs nicely for all kinds of makes use of, significantly coding. And, as Altman identified, one space the place it’s undoubtedly competing nicely is value. “Very proud of the pricing we’re in a position to ship!” he tweeted.
The highest-level GPT-5 API prices $1.25 per 1 million tokens of enter, and $10 per 1 million tokens for output (plus $0.125 per 1 million tokens for cached enter). This pricing mirrors Google’s Gemini 2.5 Professional primary subscription, which can also be in style for coding-related duties. Google, nonetheless, expenses extra if inputs/outputs cross a heavy threshold of 200,000 prompts, which means its most consumption-heavy clients find yourself paying extra.
However OpenAI is absolutely undercutting Anthropic’s Claude Opus 4.1, which begins at $15 per 1 million enter tokens and $75 per 1 million output tokens. (Anthropic does, nonetheless, supply huge reductions for immediate caching and batch processing — storing/reusing prompts and processing a number of requests collectively.)
Anthropic’s mannequin has been extraordinarily in style amongst programmers, each as a selection inside in style coding assistant Cursor and powering its personal such assistant, Claude Code. (Notice that Cursor offered GPT-5 as an option minutes after it was introduced.)
Builders who’ve had early entry to GPT-5 are touting the pricing. Simon Willison, one of many builders featured in OpenAI’s launch video, writes in his review: “The pricing is aggressively aggressive with different suppliers.” (Emphasis his.)
Techcrunch occasion
San Francisco
|
October 27-29, 2025
However GPT-5 can also be 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 implausible. Intelligence per greenback continues to extend.”
Some on X known as OpenAI’s charges for the mannequin “a pricing killer,” whereas others on Hacker Information are offering similar praise.
Will rivals like Anthropic observe? Will Google — who undercut OpenAI on pricing earlier than — get much more inexpensive? If that’s the case, we might be witnessing the beginning of a a lot awaited LLM value warfare.
There’s little question a value warfare could be welcome. The underlying economics of vibe-coding software suppliers, as an illustration, is fairly shaky due to the excessive and unpredictable charges they must pay mannequin makers, as TechCrunch’s Marina Temkin stories. And there are numerous startups constructing on high of AI fashions as nicely.
Silicon Valley has been hoping that the LLM price-to-performance ratio will ultimately enhance, together with inference prices. But it surely appeared like such an equalization might be years away because the tech business invests tons of of billions to construct information facilities and infrastructure to help rising AI demand.
OpenAI itself has a $30 billion-per-year contract with Oracle for capability, when it solely just lately hit annual recurring income of $10 billion. In the meantime, Meta plans to spend as much as $72 billion on AI infrastructure in 2025, and Alphabet has set aside $85 billion for capital expenditures in 2025, pushed by AI wants. Within the face of such huge bills, prices usually go a technique: upwards.
Given such investments, it might be too quickly for startups taking a look at their rising mannequin API payments to rejoice from OpenAI’s lone transfer to decrease pricing.
But this week, OpenAI threw down the gauntlet to place stress on costs not simply as soon as however twice. We’ll see if others observe.