Hi, we’re Maxwell 👋

Maxwell develops efficient open models for agentic work, grounded research, and software engineering.

Website · GitHub


🆕 Maxwell V1.0 Flash

A decoder-only sparse MoE model configured for agentic workflows, grounded research, software engineering, and long-context work.

35B total parameters · 3B activated parameters · 256K context window

Try in Playground · Quick Start · Technical Report

🧑‍💻 A little about us

The Maxwell team builds efficient models for agentic work: planning, tool use, code generation and analysis, and long tasks that run over many steps.

We believe cost, not intelligence, is what stops agents scaling in production. So we optimise for capability per token, per second and per dollar.

💬 Say hello

Have a question or feedback to share? Reach us at contact@maxwell-llm.ai.

Follow us here on Hugging Face for updates!