OpenAI’s latest model, announced Monday, demonstrates major strides in voice and image recognition and has the potential to significantly alter the landscape of startups building in the generative AI space, investors say.
In a livestream event Monday watched by more that 64,000 viewers, OpenAI unveiled its GPT-4o, a multimodal model with faster response times and the capacity to understand commands across voice, text, and images, translate speech between 50 languages, and perceive emotional cues from users.
“The real-time voice feels like a total game changer,” said Daniel Chesley, a principal at Work-Bench, a seed-stage enterprise fund. Chesley previously worked as a product manager at machine learning startup Hyperscience.
Since its founding in 2015, OpenAI has raised $11.3 billion in venture capital, including a $10 billion deal with Microsoft. Its launch of ChatGPT in November 2022 served as a tipping point for investment in AI, setting off a flurry of activity which sent AI startups’ valuations to record heights despite a broader slowdown in venture deals.
GPT-4o reflects the advantages offered by its unique data moat, preferential access to chips, and first-mover status, and comes just one day before Google’s flagship event for developers, Google I/O.
For enterprises which are looking to integrate generative AI technology into their back-end functions like customer support, the real-time voice capacity of GPT-4o may turbocharge that shift, VCs said. “I think we will see a refocusing of priorities within Gen AI app development towards activating users in a totally different way,” said Michael Stewart, partner at Microsoft’s CVC arm, M12.
GPT-4o’s voice capabilities have the potential to impact industries like education and healthcare, thereby setting the groundwork for new startups to build on its models.
Text-to-speech company ElevenLabs welcomed the news: “We’re excited to see companies like OpenAI join us in this space,” they told PitchBook. ElevenLabs, which has raised over $100 million in venture capital from VCs including Andreessen Horowitz, Nat Friedman and Daniel Gross, doesn’t use OpenAI’s large language model.
OpenAI’s unmatched distribution network and dataset could pose a particular threat to startups like ElevenLabs and language translation startup DeepL, both of which offer use cases similar to the tools offered by GPT-4o.
Cheaper, but still not cheap enough
OpenAI also announced new pricing Monday for developer APIs that are 50% cheaper than its most recent GPT-4 Turbo model.
But James Currier, general partner at NFX, says that OpenAI’s news doesn’t change much. “This is nothing unexpected, this is not ahead of schedule or behind, it’s right on time.” Currier said that the pricing change will make no difference and is still prohibitively high.
“If you’re doing something in the consumer space or you’ve got lots of customers, your costs are still incredibly high.”
The API news may not be enough to convince startups to stick with OpenAI, which is expensive. Enterprises are also wary of handing over their proprietary data to the company’s black box.
“We’re seeing a big trend toward lift and shift: People have experimented with OpenAI, they have something that works, and now they don’t want to pay the bill” said Joe Floyd, general partner at Emergence Capital.
Floyd said that he’s seeing many application-layer startups shift 90% of their inference calls from OpenAI’s model into open-source, fine-tuned models.
For startups still experimenting on OpenAI’s API, GPT-4o marks a significant step-change in technological capacities of multimodal models and the opportunities to build off of the tech stack.
“To me this was a true sci-fi moment for technology,” said Chesley.
Featured image of OpenAI CEO Sam Altman by Justin Sullivan/Getty Images
Learn more about our editorial standards.
