We've released version 3.0 of the ML.ENERGY Leaderboard and Benchmark!

This update includes several big updates:

• New models: OpenAI GPT OSS, Qwen 3, DeepSeek R1, and more
• New tasks: Problem solving with reasoning, and more
• Up-to-date software & hardware stack, including NVIDIA H100 and B200 GPUs
• Completely revamped web UI
• Ground-up rewritten benchmark driver that's easy to run

If you want to browse the data and feel the vibe, head to the leaderboard.

If you want to run the benchmark on other runtimes, models, or hardware, take a look at the benchmark repository.

If you want a deep technical understanding of the benchmark data so that you can interpret and explain why energy is like so, give our blog post a read. You can also read it in PDF format.