Deep Cogito Raises $43M Series A to Build Self-Improving AI Models That Autonomously Advance

Deep Cogito Secures $43M to Advance AI Self-Improvement Research

Deep Cogito has raised $43 million in Series A funding led by TQ Ventures, with participation from Benchmark, Nexus Venture Partners, Atreides Management, South Park Commons, and publicly traded cybersecurity provider Zscaler.2

2

The round brings the startup's total outside funding to over $56 million since its 2024 founding.2

2

Founded by former Google AI Search veterans Drishan Arora and Dhruv Malrana, Deep Cogito focuses on developing self-improving AI models that can autonomously advance their own intelligence without relying on human-generated training data.1

1

Groundbreaking Post-Training Techniques Drive Model Development

The startup's research centers on post-training techniques that occur after models have been trained on large volumes of general data. Deep Cogito employs Iterated Distillation and Amplification (IDA), a method where models receive additional computational resources to generate answers they couldn't produce in a single step.1

1

These improved outputs are then integrated back into the model's underlying parameters, creating a self-reinforcing cycle of improvement. When an LLM receives a prompt, IDA boosts available infrastructure to increase output quality, then analyzes this improvement to identify parameter refinements.2

2

Large-Scale Reinforcement Learning With Process Supervision

Deep Cogito's AI model training methodologies incorporate large-scale reinforcement learning enhanced by process supervision. Unlike traditional reinforcement learning that only evaluates final responses, the company's approach grades individual reasoning steps an LLM takes to reach its answer.2

2

This granular feedback mechanism makes models less likely to take unnecessary or erroneous steps, reducing hardware usage while improving accuracy. A second neural network provides feedback on task quality, helping LLMs refine their reasoning capabilities throughout the problem-solving process.2

2

Cogito Family of LLMs Demonstrates Technical Leadership

Deep Cogito released its Cogito family of open-source large language models in April 2025.2

2

The newest entry, Cogito v2.1 671B, launched in November and was described as "ahead of any other U.S. open model" available at the time.2

2

The model demonstrated superior token usage efficiency compared to comparable reasoning models, making it more cost-effective for enterprise deployment. These open-source releases showcase the practical application of the company's Infrastructure-Driven Adaptation techniques.

Enterprise Platform for Custom AI Model Development Gains Traction

The startup monetizes its technology through an enterprise platform that enables organizations to develop custom AI models using their internal data.2

2

Zscaler, one of the Series A funding participants, already uses the platform, validating Deep Cogito's commercial approach. The new capital will fuel expansion of this enterprise business while supporting research team growth and additional open-source model releases.2

2

Competitive Landscape Intensifies Around Self-Improving AI Models

Deep Cogito operates in an increasingly crowded market pursuing AI self-improvement breakthroughs. Recursive Superintelligence raised $650 million in May from investors including Nvidia, while Ineffable Intelligence closed a $1.1 billion round at a $5.1 billion valuation in April.2

2

This competitive intensity signals investor confidence that autonomous model improvement represents a critical frontier in artificial intelligence development. Watch for Deep Cogito's upcoming open-source releases and enterprise partnerships as indicators of whether its post-training methodologies can deliver sustained competitive advantages in this high-stakes race.