Fireworks AI says open models trail frontier by three months
Fireworks AI co-founder Chenyu Zhao says open models now trail frontier systems by three to six months, as the $17.5 billion startup recovers from breaches that hit Notion and 150,000 Uber drivers.
By Olivia Hart
4 min read

What's News
- Open models are at most six months behind frontier AI, probably closer to three, says Fireworks AI co-founder Chenyu Zhao
- Fireworks AI was valued at $17.5 billion by investors this summer and is reportedly in early talks for a new round
- Two fall security incidents exposed data including records on about 150,000 Uber drivers; Notion has disconnected from Fireworks
- Palo Alto Networks agreed on Wednesday to buy rival Applied Compute for $4.5 billion in stock
- Zhao says he will "probably call my broker tonight and buy a few shares" of Anthropic, which went public the same day
Open-weight AI models now lag frontier systems by three to six months, down from a gap of a year or more, according to Fireworks AI co-founder Chenyu Zhao — and he argues most enterprises no longer need to pay for the difference.
Zhao made the claim in an on-stage interview at the Slush startup conference in Helsinki, held on the same day Anthropic went public in what the source describes as the largest IPO on record. Fireworks, which runs AI models for other businesses and counts Uber, Shopify and Notion among past customers, was assigned a $17.5 billion valuation by its investors this summer and is reportedly in "early talks" for another funding round. Zhao declined to comment on those talks.
The core of Zhao's pitch: companies buy frontier intelligence they rarely use. "If you're just using Opus or Sonnet, that's not a differentiator," he said. Fireworks takes open models — which anyone can download — tunes them on a customer's own data, and delivers systems that run faster and cheaper than those of Anthropic or OpenAI and are harder for rivals to copy.
How big is the gap, really?
Zhao's answer on the open-versus-frontier question has shifted with the market. "A few years ago I'd have said a year, a year and a half," he said. "Then came DeepSeek and a waterfall of fantastic open models. These days it's at most six months behind, probably closer to three."
That is a founder's estimate, not a benchmark, and Zhao has an obvious stake in a shrinking gap. But his framing of when to switch is concrete: frontier models are low risk before product-market fit, yet once a company passes 10,000 to 100,000 users and spends $10,000 to $100,000 a month on tokens, open source becomes a rational economics question.
He also invoked Microsoft CEO Satya Nadella's July warning that proprietary-model customers pay twice — once in money, once in know-how absorbed through their prompts. Zhao predicts a "two-model world" with frontier and open models running in parallel, driving interest in routing and a shift from cost-per-token to cost-per-task thinking.
What about the security breaches?
Fireworks arrives at this moment with its first black eye. The company disclosed two security incidents this fall. In one, hackers reached API keys and tokens stored as secrets for a limited number of accounts. In the other, attackers breached Fireworks' cloud provider, entered a development system, and took customer data including billing records and some datasets.
The damage is quantified. Notion says AI chat transcripts from fewer than 1% of its workspaces were exposed — and it has disconnected from Fireworks. Uber says hackers stole names, addresses, birth dates and driver's license numbers of about 150,000 drivers through a vendor TechCrunch identified as Fireworks.
Zhao defended the company's record. "Every company is probably seeing an unprecedented frequency of attacks," he said, attributing much of the motivation to "LLM jacking" — hunting API keys to sell tokens on illicit marketplaces. "Our promise is zero data retention for inference traffic, and there's no indication that has been compromised."
The Notion exposure, he explained, came through batch prediction, where uploaded datasets are stored at rest — "it's not a zero-retention product by nature." Fireworks is moving those services to an online flow that keeps data only as long as needed. He declined to name other affected customers beyond Notion and Cursor, calling it "a small subset."
Is the money still flowing?
Competition and capital are crowding the category. Rivals Together AI and Baseten have announced large rounds in recent months. On Wednesday, Palo Alto Networks said it will buy Applied Compute — a startup founded a year ago by former OpenAI researchers that, like Fireworks, helps companies train and run custom models on their own data — for $4.5 billion, all in stock.
Even Zhao hedges his own thesis. Asked about Anthropic's freshly public valuation, he said: "On the record: I'll probably call my broker tonight and buy a few shares. This is a long-term bet." He argued AI usage in 10 years will operate at a completely different scale, requiring the enormous capital that public markets supply.
The open-model bet, in short, is not a bet against frontier AI — it is a bet that both can win at once. Fireworks' challenge is proving that cheap, controllable intelligence can survive contact with both hackers and a $4.5 billion consolidation wave.
Original: fireworks.ai
More from Olivia Hart
Show full bio
Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,450 articles