Meta Releases Llama 4 Under Permissive Open-Source License
Meta's latest open-weight model matches proprietary competitors on key benchmarks while remaining freely available to researchers and developers.
Open Source Catches Up
Meta has released Llama 4, the latest in its open-weight language model family. Available under the Llama Community License, the model can be used commercially with minimal restrictions, marking a significant moment for the open-source AI ecosystem.
Performance Highlights
Llama 4 closes the gap with proprietary frontier models:
- Coding: Matches GPT-5 on HumanEval and SWE-bench
- Reasoning: Within 3% of Claude on graduate-level math and science
- Multilingual: Supports 40 languages with strong performance in low-resource settings
Model Variants
Meta is releasing three sizes: Llama 4 Scout (8B), Llama 4 Pro (70B), and Llama 4 Maverick (405B). All variants include a 256K token context window.
Community Response
The Hugging Face community has already begun fine-tuning Llama 4 for specialized applications. Within 48 hours of release, over 200 derivative models appeared on the platform.
Implications for the Industry
The release intensifies the debate over open vs. closed AI development. Proponents argue that open weights democratize access and enable safety research; critics worry about misuse potential without API-level controls.
