Meta has released Llama 4, the latest version of its open-source large language model family, which the company claims rivals or exceeds GPT-4 on standard benchmarks while being freely available for commercial use. The release marks a significant moment in the AI democratization debate and puts additional pressure on closed-model providers like OpenAI and Anthropic.
What Makes Llama 4 Different
Llama 4 introduces a mixture-of-experts architecture that dramatically improves efficiency without sacrificing capability. The model comes in multiple sizes — 8B, 70B, and 400B parameters — with the largest variant delivering performance that Meta says matches or beats GPT-4 Turbo on coding, reasoning, and instruction-following tasks. Crucially, the models are released under a permissive license that allows commercial use for most applications.
Benchmark Performance
On MMLU, Llama 4 70B scores 86.1%, compared to GPT-4’s 86.4% — essentially equivalent performance at a fraction of the cost to run. On HumanEval coding benchmarks, the 70B model scores 81.7%, outperforming the previous generation by 12 percentage points. Meta’s internal evals show strong performance on multilingual tasks, which has historically been a weakness of open models.
Community and Commercial Impact
The open-source AI community has responded enthusiastically. Within 24 hours of release, Llama 4 models had been downloaded over 500,000 times from Hugging Face and fine-tuned versions were already appearing for specialized domains including medical, legal, and financial applications. Developers praised the permissive licensing compared to earlier versions that restricted certain commercial uses.
For businesses, Llama 4 offers a compelling alternative to paid API services: a company can run a capable AI model on its own infrastructure for the cost of cloud compute, with no per-token fees and complete data privacy. This proposition is particularly attractive for enterprises handling sensitive data.
Zuckerberg’s Open-Source Bet
Meta CEO Mark Zuckerberg has been outspoken about his belief that open-source AI is the right strategy, both ethically and competitively. “A world where AI is controlled by a small number of closed-model companies is not good for anyone,” Zuckerberg wrote in a post accompanying the release. Analysts note that open-sourcing AI models also benefits Meta indirectly by establishing its preferred AI standards and undermining competitors’ pricing power.
