Case Study Deskcasestudydesk.com
AI Search / AgentsSourced

Case Study: Genspark cut cost 50% and shipped in 4 weeks with Fireworks AI

Genspark Case StudySourced & dated by Case Study Desk
Key facts · TL;DR
Company
Genspark
Industry
AI Search / Agents
Challenge
Needed better quality and lower cost than a closed model
Headline result
Post-training on Kimi-K2 beat a closed model on quality at half the cost

Key results

50%
Cost reduction
vs closed-source model
12%
Higher quality
vs SOTA closed model
33%
More tool calls
5 vs 3.74
4 wk
POC to production

The challenge

Genspark wanted to own its AI stack and improve output quality while reducing reliance on a proprietary closed-source model, without a long development cycle.

The solution

Genspark used Fireworks AI to post-train and deploy a Kimi-K2-based model, moving from proof of concept to production in four weeks.

“Fireworks enabled us to own our AI journey, and unlock better quality in just four weeks. This resulted in a better user experience for our customers.”

KZ
Kay Zhu
CTO, Genspark

The results, in context

Genspark reported a 50% cost reduction versus a closed-source model along with 12% better quality and 33% more tool calls (5 versus 3.74 for the proprietary model). The company moved from POC to production in four weeks.

Products used

Fireworks AI Fireworks AIFireworks AI Fine-tuning / post-training