AI development startup Mirendil Inc. has raised $1 billion in Series A funding co-led by Kleiner Perkins and Clearlake Capital, with participation from Google. Founded last year by former Google DeepMind and Anthropic researchers Behnam Neyshabur and Harsh Mehta, the startup develops software designed to automate complex machine learning pre-training workflows.

Mirendil’s flagship platform, Lor, coordinates hundreds of autonomous AI agents to automate model training, evaluation, and error correction. To demonstrate the software, a single researcher built an open-source competitor model called Mirendil-Base in 3.5 months for $10 million, with agents writing over 99 percent of the underlying code, provisioning infrastructure, and curating web training datasets without team assistance.

The company recently leveraged its technology to launch Fern, a specialized programming model offering a 10-million-token context window designed to tackle automated software engineering. Mirendil plans to deploy its fresh capital toward building larger base models ranging up to 10 trillion parameters.

Why it matters

  • Shows massive venture capital appetite for agentic platforms that dramatically lower the cost and headcount required to train foundation models.

  • Demonstrates agent-driven automated workflow capabilities that scale token context limits to 10M tokens for developer workloads.

  • Signals intense competition for pre-training efficiency tools that reduce trial-and-error costs for enterprise AI labs.

Source: siliconangle.com