Deep Cogito Inc., an AI startup founded by former Google engineers, has raised $43 million in a Series A round led by TQ Ventures. The funding brings the company's total outside capital to more than $56 million as it works on AI models that can improve themselves.
The round included Benchmark, Nexus Venture Partners, Atreides Management, South Park Commons, and Zscaler Inc., a publicly traded cybersecurity provider. Deep Cogito plans to use the money to grow its enterprise business, expand its research team, and release new open-source models.
Self-improving models
Deep Cogito was founded in 2024 by Drishan Arora and Dhruv Malrana, both former Google employees. The company's research has produced Cogito, a family of open-source large language models first released in April 2025. The latest version, Cogito v2.1 671B, debuted in November and the company said it was "ahead of any other U.S. open model" available at the time. The algorithm also used less token usage than comparable reasoning models, a measure of LLM cost-efficiency.
The company trains its models using a technique called IDA. When given a prompt, IDA increases the computing infrastructure available to the LLM to boost output quality, then analyzes the improvement to identify ways to refine the model's parameters. This approach is central to the company's work on Generative AI and LLM development.
Training methodology
Deep Cogito also uses reinforcement learning, where an LLM receives sample tasks and gets feedback on its responses from a second neural network. That feedback helps the model refine its reasoning. Most reinforcement learning workflows only grade an LLM's final answers, but Deep Cogito uses a method called process supervision to evaluate each step the model takes to reach a response. According to the company, this makes its models less likely to take unnecessary steps, which reduces hardware costs.
The company monetizes its technology through a platform that lets enterprises build custom AI models using their internal data. Zscaler, one of the investors in the round, is already using the platform. For AI for IT & Development teams, the company's approach to reducing token usage and hardware requirements could matter when evaluating model costs at scale.
Competitive landscape
Deep Cogito is one of several startups pursuing self-improving AI models. Recursive Superintelligence Inc. raised $650 million in May with backing from Nvidia Corp. In April, Ineffable Intelligence Ltd. closed a $1.1 billion round at a $5.1 billion valuation.
Why this matters for IT and development teams
Process supervision and IDA-style training directly address two costs that matter in production: token consumption and compute usage. If Deep Cogito's open-source models continue to deliver comparable reasoning quality with lower infrastructure requirements, development teams evaluating self-hosted LLM options may have a new benchmark to measure against. The company's enterprise platform also gives IT teams a path to customize models on internal data without starting from scratch.
Your membership also unlocks:
