Pulse News

**OpenAI’s GPT-4.5 Underwhelms; Is Poolside Worth $5 Billion?**

by Stephanie Palazzolo, Aaron Holmes, and Sri Muppidi

Mar 3, 2025, 7:00am PST

Poolside co-founder Jason Warner. Photo via Getty.

OpenAI must have been feeling the pressure from Anthropic and XAI's recent model releases because on Thursday it launched GPT-4.5 (Orion), its latest large language model. But OpenAI went out of its way to lower expectations for the model.

On X, CEO Sam Altman described GPT-4.5 as a "giant, expensive model" that "won't crush benchmarks." By OpenAI's own admission, the new model generally performs worse than a number of other models, such as Anthropic's just-launched Claude 3.7 Sonnet and even some of OpenAI's own reasoning models that have been available for months. It's also incredibly expensive, costing more than 10 times more than 3.7 Sonnet.

The release received largely unenthusiastic responses on X from developers, though some felt it was better than GPT-4.0, which came out in May last year, for some queries. While GPT-4.5 has redeeming qualities such as more realistic dialogue and a better sense of humor, which could be useful for voice interactions, the model's performance is clearly nowhere near what the company had originally hoped for. It looks like evidence of slowing gains from pretraining, which we first told you about in November.

The question now is when will OpenAI launch what it calls GPT-5, combined with its 03 reasoning model.

**In other developments...**

Even as OpenAI and Anthropic increasingly focus on artificial intelligence products for software engineers, venture capitalists are constantly funding would-be challengers at sky-high prices. In recent weeks, for instance, investors have been talking to executives at Poolside, a two-year-old AI coding developer founded by one of the creators of GitHub's Copilot coding assistant, about investing more in the company, according to three people who spoke to Poolside leaders.

The company could target a $5 billion valuation in an eventual fundraising, according to a person who recently spoke to Poolside CEO Jason Warner. (For its part, Poolside isn't interested in raising new financing now and hasn't been discussing a potential valuation, a person close to the company said.) A $5 billion valuation would be striking given that Poolside raised $500 million at a $3 billion valuation in a deal led by Bain Capital Ventures just four months ago. And the company generated less than $10 million in revenue in 2024, according to two of the people who spoke to its executives, though we don't know what it's projecting this year.

Poolside is developing a coding assistant app and two coding-focused models—one for more complex engineering tasks and the other for simpler code completion. Though the products aren't available broadly to developers yet, they will likely compete with similar tools from OpenAI and Anthropic. Investors such as Vinod Khosla, OpenAI's first venture investor, are also looking to back startups developing software that essentially aim to replace engineers entirely, such as coding agent Devin, he said at The Information's AI Agenda Live event last week.

OpenAI has been developing a similar product to Devin. And much of ChatGPT's subscription revenue comes from software engineers who use the chatbot for coding tasks. That's one of the reasons investors are going ga-ga for the coding sector.

Even if Poolside hypothetically hoped to generate $100 million in revenue this year, a $5 billion valuation would still rank it among the most expensive generative AI startups to have raised money recently, in terms of valuation multiples. Even at its existing $3 billion valuation, Poolside looks pricey compared to the three-year-old startup behind AI coding assistant Cursor, which recently hit $100 million in annual recurring revenue—subscription revenue it could generate over the next 12 months—and it has raised money at a $2.5 billion valuation. Codeium, another AI coding assistant startup, is in talks to raise funding at a nearly $3 billion valuation in a round led by Kleiner Perkins, though we don't have its revenue information.

**Homegrown Models**

Poolside may get some help from Amazon. In December, the startup announced a deal with Amazon Web Services to sell its coding models—Malibu and Point—as well as its coding product. As part of the partnership, AWS will sell Poolside's product as a first-party service—the cloud giant will make the startup's product available to customers alongside AWS' baseline web services like storage and compute—which Poolside expects will help grow its revenue significantly in the coming years, according to one of the people who spoke to the startup.

To sell its product, Poolside is pitching senior executives and decision makers at large companies, according to one of the people. Poolside's product is tailored to big enterprises: its models are customized on customers' codebases. Customers can run Poolside's models on their own servers instead of in the cloud. That's an important consideration for defense contractors, for instance.

Poolside's strategy differs from the approach pursued by other AI coding firms, including the makers of Cursor and Devin, which have gained traction with individual software developers. Unlike many other AI coding-related firms such as the maker of Cursor and even Microsoft's GitHub Copilot, Poolside doesn't rely on models from Anthropic and OpenAI. Poolside will have to keep its models competitive with those of the incumbents but some investors may prefer that challenge over the risks of relying on competitors.