Meta Open-Sources Llama 3.1 405B. It Matches GPT-4. The Generosity Is Strategic, Obviously.
Meta released Llama 3.1 405B, an open-weights model that reportedly matches GPT-4 level performance. The model is runnable on consumer GPUs, meaning developers and small companies can build frontier-grade applications without paying API fees or hitting usage ceilings. The weights are publicly available.
This illustrates the concept of capability diffusion, where frontier-level intelligence escapes the paywall and becomes infrastructure. The mechanism is open-weights release: Meta trains the model, foots the compute bill, and then distributes the parameters for anyone to run locally. The lesson is that the moat around frontier AI is shrinking. If you are paying per token for capabilities that now exist in open models, you should be asking harder questions about your architecture.
Meta built and released Llama 3.1 405B. They claim it is the first open model that rivals closed frontier models and that it runs on consumer GPUs. Per their announcement, developers can build GPT-4 level apps without API costs or usage limits.
- Visit ai.meta.com and find the Llama 3.1 download page or the Hugging Face model card linked from Meta's blog post. Review the hardware requirements before you get your hopes up, as 405 billion parameters is not trivial.
- Download a smaller Llama 3.1 variant if your GPU cannot handle the full 405B model. The smaller variants still demonstrate the same open-weights principle and will run on more modest hardware.
- Use a local model runner like LM Studio or Ollama to load the model and chat with it offline. The expected outcome is a fully local, private conversation with a frontier-grade model and zero API charges, which should make the economics of closed APIs suddenly feel less compelling.