Meta's Llama 4 Outperforms Commercial Models While Remaining Open Source

The freely available model achieves state-of-the-art results on multiple benchmarks, challenging the business models of closed-source AI companies.

Meta has released Llama 4, the latest version of its open-source large language model, which surprisingly outperforms several leading commercial models on standard benchmarks while remaining freely available for research and commercial applications.

The new model, available in 8B, 70B, and 400B parameter versions, scored higher than GPT-4 and Claude 3 Opus on the MMLU (Massive Multitask Language Understanding) benchmark, achieving a score of 92.3% on this comprehensive test of knowledge and reasoning.

"We believe in democratizing access to advanced AI," said Mark Zuckerberg in a blog post announcing the release. "Llama 4 represents our commitment to open science while pushing the boundaries of what's possible with publicly available models."

The Verge tested the 70B parameter version against several commercial alternatives and found it matched or exceeded their performance on coding tasks, creative writing, and factual accuracy. The 400B parameter version, which requires significant computational resources to run, demonstrated even more impressive capabilities, particularly in complex reasoning and specialized knowledge domains.

The model is being released under a permissive license that allows commercial use, though with certain safety restrictions. This move has been praised by the open-source community but has raised concerns among some AI safety researchers who worry about potential misuse.

Meta's decision to release such a capable model for free poses significant challenges to companies like OpenAI and Anthropic, whose business models rely on charging for access to their most advanced models. Industry analysts suggest this could accelerate the commoditization of foundation models, pushing companies to differentiate through specialized applications and services rather than the base models themselves.

"This is potentially disruptive to the current AI business landscape," said Dr. Jim Fan, an AI researcher at NVIDIA. "If open-source models continue to close the gap with or even surpass proprietary ones, it fundamentally changes the economics of the industry."

Meta also announced a partnership program to help startups and researchers leverage Llama 4 for their applications, including access to specialized fine-tuning resources and technical support.

The release has reignited debates about AI openness, with proponents arguing that open models accelerate innovation and democratize access, while critics express concerns about reduced safety oversight and potential misuse. Meta maintains that their responsible disclosure approach, which included pre-release briefings with safety researchers and policymakers, strikes the right balance between openness and safety.