Open-weight AI models within six months, Fireworks co-founder says
Fireworks co-founder Chenyu Zhao says open-weight models trail frontier systems by three to six months, as enterprise adoption rises and two security incidents strain trust.

- Zhao estimated open-weight models are at most six months behind frontier systems, probably closer to three.
- Fireworks investors valued the company at $17.5 billion in summer.
- Notion reported exposure in fewer than 1% of workspaces; Uber reported data affecting about 150,000 drivers.
- Palo Alto Networks said Wednesday it would buy Applied Compute for $4.5 billion in stock.
- Zhao said companies should evaluate open models at 10,000 or 100,000 users and $10,000-$100,000 in monthly token spending.
By Nathan Brooks
Open-weight AI models now trail frontier systems by no more than six months, Fireworks AI co-founder Chenyu Zhao said at Slush in Helsinki.
Zhao put the likely gap at three months. A few years earlier, he would have estimated one to 1.5 years before DeepSeek and a rapid wave of open models. His estimate remains a founder's judgment; Zhao supplied no independent benchmark.
On the same day Anthropic went public in the largest IPO on record, investor enthusiasm for the AI lab ran high. At the Slush startup conference in Helsinki, the central theme was open-source everything. Zhao believes enterprises can usually avoid frontier prices unless they need the latest and greatest generalized intelligence.
Zhao spent a decade at Google, including managing a 50-person engineering team on Google Cloud Vertex AI. He said he left after meeting co-founders Lin Qiao and Dima Dzhulgakov, whom he judged smart within five minutes and genuinely nice.
Why do enterprises consider open-weight models?
Fireworks runs AI models for other businesses and specializes in downloadable open-weight models. Customers can tune them on their own data, which Zhao says produces systems that run faster and more cheaply than Anthropic, OpenAI and other frontier models, while making them harder for competitors to copy. Uber, Shopify and Notion have used Fireworks.
Zhao said: "We use open models, which are often smaller, cheaper and faster, and let companies add their own data to specialize them. That way you own your AI rather than being beholden to giant AI conglomerates."
He described Fireworks as a specialized intelligence platform, while OpenAI and Anthropic pursue general intelligence. For a company using only Opus or Sonnet, he said, Fireworks offers no differentiator.
Frontier models remain household brands, Zhao said, so selecting the strongest system carries little risk before product-market fit. At 10,000 or 100,000 users, with monthly token spending of $10,000 to $100,000, he said, companies should evaluate open source.
In July, Microsoft CEO Satya Nadella warned that proprietary models make companies pay twice: once in money and again in accumulated know-how. Frontier labs can also absorb customer knowledge through the prompts used to obtain answers.
Zhao expects a two-model market, with frontier and open models operating in parallel. Frontier systems can handle high-level planning, but basic arithmetic does not require one. That distinction explains interest in routing and a shift from cost per token to cost per task.
What do Fireworks' security incidents expose?
Fireworks 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.
Notion said hackers exposed AI chat transcripts from fewer than 1% of its workspaces, and it disconnected from Fireworks. Uber said hackers stole names, addresses, birth dates and driver's license numbers from about 150,000 drivers through a vendor that TechCrunch identified as Fireworks.
Zhao would not name other affected customers. He said Notion and Cursor had already disclosed their impact, and he described the affected group as a small subset of Fireworks' customers.
Zhao said every company was probably seeing an unprecedented frequency of attacks. He described the motivation as LLM jacking: searching providers for API keys and secrets to sell tokens on illicit marketplaces. Fireworks promises zero data retention for inference traffic, and Zhao said no evidence showed that protection had been compromised.
Notion used batch prediction, which stores uploaded datasets at rest, so the product cannot be zero-retention by nature. Zhao said Fireworks is moving those services to an online flow that keeps data only as long as needed.
How is capital moving around enterprise AI?
Fireworks investors valued the company at $17.5 billion in summer. It reportedly entered early talks for another funding round, which Zhao declined to comment on.
- Together AI and Baseten announced large funding rounds in recent months.
- On Wednesday, Palo Alto Networks said it would buy Applied Compute for $4.5 billion, all in stock. Founded a year earlier by former OpenAI researchers, Applied Compute helps companies train and run custom AI models using their own data.
Nvidia has backed Fireworks since 2024. After Jensen Huang left Nvidia the previous week, Zhao called the departure a blow to the industry and expressed sympathy for Huang and his family. Zhao expects Nvidia to keep investing in Nemotron and maintain its Fireworks partnership, while noting that Nvidia has always supported open source.
What does Anthropic's IPO mean for Fireworks?
Asked about Anthropic's public-market valuation, Zhao described two possible paths. Anthropic could cement its position in an AI oligopoly, or it could lose traction as open models improve.
On the record, Zhao said: "I'll probably call my broker tonight and buy a few shares." He called the move a long-term bet. In 10 years, he expects AI usage to reach a completely different scale, requiring substantial capital and helping explain why model companies are going public.
Monitor whether routing shifts enterprise budgets from cost per token to cost per task, whether Fireworks retains customers after two disclosures, and whether its reported funding talks and Palo Alto Networks' $4.5 billion stock purchase close. Anthropic's public trajectory will test Zhao's oligopoly-versus-losing-traction forecast.
via techcrunch.com (Source)
More from Nathan Brooks
Show full bio
Senior reporter covering companies, deals and strategy in Search rankings and algorithm updates, generative engine optimization at SERP Journal.
226 articles