Meta has released Llama 4, its most capable open-source AI model, which benchmarks show performing within 5% of GPT-5 on most tasks while being freely available for commercial use.
Performance
Llama 4 matches GPT-5 on coding (97% on HumanEval), surpasses it on multilingual tasks, and comes close on reasoning benchmarks. The model can run on consumer-grade hardware with quantization.
Market Impact
The release threatens OpenAI's business model by offering near-equivalent capabilities for free. Enterprise adoption of open-source AI models has doubled since the announcement.
- Within 5% of GPT-5 on most benchmarks
- Free for commercial use
- Runs on consumer hardware
- Enterprise open-source adoption doubled
