OpenLake launches fast inference for self-improving models
OpenLake announced its platform on Y Combinator's Launch YC. It says it lets companies deploy, train and serve models that keep improving, automatically post-training them as new data arrives.
The company says you pick an open model such as DeepSeek V4 or GLM 5.3 and deploy it on OpenLake, plug in datasets, RL environments and data sources, and it post-trains the model continuously and serves the improved version in place, with A/B tests against previous versions. OpenLake says trained checkpoints belong to the customer and that on average its models cut serving costs in half.
Would you let your models keep training themselves? Send this to someone running models in production. The launch post says OpenLake is available today at openlakehq.com. Founder Arnav Balyan, whom Y Combinator lists as the founder (Founder/CEO), says he previously built core infrastructure behind Uber's data and AI systems.
The story in cards






Sources: OpenLake on Launch YC and YC company page. Photo: ycombinator.com/companies/openlake-technologies.
