Fireworks AI says firms can skip pricey frontier models, as Anthropic lists
At Helsinki's Slush conference, Fireworks AI argued open-weight models suit most business needs, even as Anthropic's record IPO dominated the day.

Fireworks AI, a startup that hosts and customizes open-weight AI models for other companies, says most enterprises do not need the costly frontier systems sold by Anthropic and OpenAI, according to TechCrunch. Co-founder Chenyu Zhao made the case on stage at the Slush startup conference in Helsinki, where open-source tools drew heavy attention on the same day Anthropic completed the biggest public listing ever recorded.
Zhao argues that open models, which anyone can download, often run smaller, quicker and more cheaply than frontier offerings. Fireworks tunes those models on a client's own data, which he says makes them faster and harder for competitors to replicate. He told the audience that unless a business requires the newest and most capable general-purpose intelligence, premium models are largely avoidable. He also said the gap between open and frontier systems has narrowed sharply, from roughly a year and a half before DeepSeek arrived to perhaps three to six months now.
The company has counted Uber, Shopify and Notion among its customers. Zhao, who spent ten years at Google and led a 50-person engineering group on Cloud Vertex AI, started Fireworks with Lin Qiao and Dima Dzhulgakov. Investors valued the business at $17.5 billion in the summer, and it is reportedly sounding out backers for a further round, a matter Zhao would not discuss.
Fireworks has also reported two security breaches this autumn. In the first, intruders obtained API keys and tokens kept as credentials for a small number of accounts. In the second, attackers got into a development machine through the company's cloud provider and took client information, among it billing details and certain datasets. Notion said conversations from under one percent of its workspaces were affected and has since cut ties with Fireworks. Uber said roughly 150,000 drivers had personal details, including licence numbers, taken via a vendor that TechCrunch identified as Fireworks.
Money continues to flow into the sector. Together AI and Baseten have each announced large funding rounds recently. Palo Alto Networks said on Wednesday it would acquire Applied Compute, a year-old startup founded by former OpenAI researchers that builds custom models on customer data, for $4.5 billion in stock.
Based on reporting by TechCrunch. RoxMag adds plain-English context and checks every figure against that source.
fireworks aiopen-weight modelsanthropicslushai security
More in AI

Maryland students sue Omnilert after AI flags Doritos bag as gun
Four students detained by armed police after an AI system misread a chips bag as a firearm have filed a lawsuit against the vendor, police and school officials.

Nadella: Microsoft Buyouts Gave 60,000 Staff a Choice as AI Reshapes Jobs
Microsoft's CEO says the October buyout offer was about giving employees agency as the company reorganizes around AI.

UK MPs Back AI Watchdog With Power to Halt Dangerous Models
A parliamentary committee wants a regulator able to stop high-risk AI training and block releases, saying firms should not police themselves.

Nvidia Says Open Source AI Push Continues After Huang Exit
Interim CEO Colette Kress told analysts Nvidia remains committed to open models and that the $12.9 billion Hugging Face deal is still on track.