Martian built a patent-pending LLM router that searches across many models to route each prompt to the LLM expected to deliver the highest quality at the lowest cost. The router also improves reliability by automatically rerouting prompts when a model or provider becomes unavailable. Accenture has invested in Martian through Accenture Ventures and plans to integrate Martian’s dynamic routing into its switchboard services to customize and control LLM combinations for clients. Martian was previously backed by $9M from NEA, General Catalyst, CVP, and Prosus Ventures. Engineers at 300+ companies, including Amazon and Zapier, have used Martian to achieve higher performance, lower costs, and greater security and reliability. The team includes former AI researchers from Stanford, Harvard, the University of Pennsylvania, the Google Bard team, and Microsoft Research. Martian builds a “model router” that automatically routes prompts to the best large language model for a given request, selecting by uptime, skillset and cost-to-performance. The router estimates how models will perform without actually running them and indexes new models with zero manual work. Founders Shriyash Upadhyay and Etan Ginsberg, AI researchers from the University of Pennsylvania, launched the company to align product development with interpretability research. Martian emerged from stealth with $9 million in funding and currently has a 10-person team; the proceeds are earmarked for product development, research into models' internal operations, and team growth. The product aims to lower costs by routing cheaper models for requests that perform similarly to pricier models and using expensive models only when necessary. Martian says it has seen some uptake, including among “multibillion-dollar” companies. Its competitive outlook will hinge on pricing and the ability to deliver in high-stakes commercial scenarios, and at least one rival, Credal, offers a similar automatic model-switching tool.

Total raised

$9M

Funding rounds

2

Latest round

Seed

Latest activity

Nov 2023