Meta Drops 405 Billion Parameters for Free
Meta released Llama 3.1 405B as open weights under a permissive license. The model matches GPT-4 on standard benchmarks and supports full fine-tuning or inference on local hardware or low-cost cloud GPUs without token fees.
Users stop treating frontier models as paid APIs and start treating them as local infrastructure. This shifts budgeting from usage costs to hardware and electricity planning.
Hugging Face hosts the weights and reports over 2 million downloads in the first week. Independent labs such as Together AI now offer hosted inference at roughly one tenth the price of closed frontier APIs.
Step 1: Visit huggingface.co/meta-llama/Meta-Llama-3.1-405B and accept the license. Step 2: Run the provided transformers script with 8-bit quantization on an A100 80 GB instance. Step 3: Observe generation speed of 18 tokens per second and zero per-token billing.