Meta Releases Llama 3.1: Open-Source Scaring OpenAI

Meta just dropped Llama 3.1, the largest open-source model ever: 405 billion parameters, commercial license, free for anyone with enough RAM.

The performance? Close to GPT-4 on many benchmarks, sometimes even better. The difference is you can download it, tweak it, host it on your servers, and nobody asks for a monthly subscription.

The catch? You need serious hardware: at least 8 enterprise GPUs to run it decently. Not laptop stuff. But for companies and researchers it's a blessing: finally a powerful model that doesn't depend on paid APIs.

OpenAI isn't commenting, but the silence says it all.