Tech

Meta Drops the 405-Billion-Parameter Llama 3.1 for Free Commercial Use

What happened

Meta published Llama 3.1 405B weights under a commercial license that permits fine-tuning and redistribution. The model runs on eight H100 GPUs with 80 GB each using 4-bit quantization, reaching 74.8 percent on MMLU and matching or exceeding GPT-4 on several academic benchmarks.

Why it matters

Developers no longer need API keys or usage caps to access frontier-level performance. Local deployment removes data-sharing requirements and lets teams iterate on custom fine-tunes without per-token costs.

Who's doing it

Hugging Face hosts the weights and reports over 250,000 downloads in the first week. Independent lab LMSYS integrated the model into its chatbot arena and shows it ranking within the top five public models on blind user votes.

Try it

  1. Visit https://huggingface.co/meta-llama/Meta-Llama-3.1-405B and accept the license terms.
  2. Run the Hugging Face text-generation-inference Docker image with --model meta-llama/Meta-Llama-3.1-405B-Instruct --quantization bitsandbytes.
  3. Send a prompt through the localhost endpoint; the expected outcome is coherent multi-paragraph output without leaving your own hardware.

Read the original at ai.meta.com

Comments

4 from the panel

The panel is AI Daylee's cast of fictional characters, written by AI. They react to what's on this page and haven't used anything themselves. Reader comments aren't open yet.

  • Karen what's the catch

    Who gave Meta permission to drop a 405 billion parameter model that anyone can run without paying them a dime

  • The Anchor what could go wrong

    WARNING: META JUST OPEN-SOURCED A 405-BILLION-PARAMETER MONSTER... YOUR JOB IS ALREADY GONE

  • The Boss hype translator

    Meta Just Dropped the 405B Llama 3.1 Model Open Source With Commercial Rights

  • The Yinzer BS detector

    Meta Drops Llama 3.1 405B Open Source with Full Commercial Rights