Kailash Labs
ActiveAI-nativeA factory for small and specialized video reasoning models
What it does
The world generates petabytes of video every day, and almost none of it is readable to machines. Frontier VLMs that can unlock this (like Gemini) are too expensive to run at the volume real industries have, too generic for any single task, and can't be deployed on-prem where sensitive video lives. We are building an auto vision factory: take video plus a customer's objective, distill a frontier model, and post-train a small open model that is SOTA on that one task and serve it blazingly fast. Our first product, Marlin, is a 2B-parameter open-weights video VLM that matches Gemini-2.5-flash on dense captioning and temporal grounding at a fraction of the cost.
Its site says now · captured 2026-09-19
Kailash Labs
Kailash Labs is a video research and product lab. We are on a mission to convert the world's videos into structured data that machines can reason over. Our aim is to build small, grounded video understanding models that can be adapted to any downstream video analysis task with little data. Towards a world where machines can watch and understand videos as fluently as humans.
Next to it in AI infra and compute · Video model distillation and serving
- HopperY Combinator F26
Fast inference for voice
- BlyssY Combinator W23
End-to-end encrypted AI.
- OrchestraY Combinator S26
Self-optimizing neocloud that cuts LLM costs by 100x
- ZantPlug and Play PnP 2026
Zant simplifies and accelerates the deployment of AI models on edge devices, where cloud-based execution is limited by privacy, latency, or connectivity constraints.
- Banbury Roada16z speedrun SR007
Banbury Road is a research lab training civilizations of models, not one bigger model.
- ConiferY Combinator S26
Least cost routing system to reduce 70%+ token spend
- PlayAIY Combinator W23
Our mission is to make Voice AI accessible and useful to all.
- PoplarMLY Combinator W23
Deploy Machine Learning Models with One Command