Deep Cogito raises $43M to develop self-improving AI models

TQ Ventures led the round, with participation from Benchmark, Nexus Venture Partners, Atreides Management, South Park Commons and Zscaler Inc., a publicly traded cybersecurity provider. The investment brings Deep Cogito’s total outside funding to more than $56 million.

Deep Cogito was founded in 2024 by former Google employees Drishan Arora and Dhruv Malrana. The company’s research has produced Cogito, a family of open-source large language models first released in April 2025. The newest version, Cogito v2.1 671B, debuted last November and was described in a company blog post as “ahead of any other U.S. open model” at the time, with lower token usage than comparable reasoning models.

The startup trains its models with a proprietary technique called IDA, which increases the infrastructure available to a model for a given prompt and analyzes the resulting output quality gains to refine the model’s parameters. It also uses reinforcement learning with a method called process supervision, which grades individual steps in a model’s reasoning rather than only the final response. Deep Cogito says this approach reduces unnecessary or erroneous steps and lowers hardware usage.

Deep Cogito generates revenue through a platform that lets enterprises build custom AI models using their internal data. Zscaler, one of the round’s investors, is a user of the platform. The company plans to use the new funding to expand its enterprise business, grow its research team and release additional open-source models.

Deep Cogito is one of several startups pursuing self-improving AI. Recursive Superintelligence Inc. raised $650 million in May from a group including Nvidia Corp., and Ineffable Intelligence Ltd. closed a $1.1 billion round in April at a $5.1 billion valuation.