From TechCrunch · Connie Loizos · · 3m
Fireworks AI pitches open models as cheaper alternative
As Anthropic goes public, a rival argues companies can skip pricey frontier AI models and use cheaper open-source alternatives instead.
In brief
As Anthropic goes public, a rival argues companies can skip pricey frontier AI models and use cheaper open-source alternatives instead. Fireworks AI argues most companies can use cheaper open-source models instead of expensive frontier AI, though recent security breaches have cost it customers. Originally reported by…
Anthropic IPO meets open-source push
On the day Anthropic went public in the largest IPO on record, Fireworks AI was pitching the opposite approach at Slush in Helsinki: open-source models.
Fireworks AI's business model
The four-year-old Bay Area company runs AI models for other businesses, specializing in open-weight models that customers can download and tune on their own data.
The case for open models
Fireworks co-founder Chenyu Zhao argues frontier models are pricey and unnecessary for most companies unless they need the latest generalized intelligence.
Data privacy concerns
Companies worry frontier labs absorb too much customer know-how through prompts. Microsoft CEO Satya Nadella raised this concern in July.
Open models closing the gap
Zhao says open models lag frontier models by at most six months, probably closer to three, after DeepSeek and other releases improved the category.
Two-model future
Zhao expects companies will use frontier models for high-level planning but open models for basic tasks, shifting focus to cost per task rather than cost per token.
Security breaches hit Fireworks
Fireworks disclosed two security incidents this fall. Hackers accessed API keys and tokens for some accounts, and breached its cloud provider to steal customer data.
Customers disconnect after breaches
Notion says fewer than one percent of its workspaces had AI chat transcripts exposed and has disconnected from Fireworks.
Fireworks rebuilds trust
Zhao says a small subset of customers were impacted by data exposure. Fireworks promises zero data retention for inference traffic and is moving batch prediction services to online flow to keep data only as long…
The big picture
Fireworks AI argues most companies can use cheaper open-source models instead of expensive frontier AI, though recent security breaches have cost it customers.
Read the TL;DR — this story on one card.




