Meta releases Llama 4 — open-source model matches GPT-4 on benchmarks
Meta's Llama 4 family achieves GPT-4-class performance across coding, reasoning, and multilingual tasks while remaining fully open-source with commercial licensing.
Original sourceLlama 4 is Meta's most capable open-source model, closing the gap with frontier closed models. The 405B parameter model matches GPT-4 on most benchmarks and exceeds it on multilingual tasks.
The family includes 8B, 70B, and 405B variants, plus a new Llama 4 Scout model optimized for on-device inference on smartphones. All models are available for commercial use under Meta's open license.
For the open-source AI community, this is validation that open models can compete at the frontier. Running Llama 4 405B requires significant infrastructure, but the smaller variants are practical for most applications.
Panel Takes
The Builder
Developer Perspective
“Llama 4 70B is the sweet spot — runs on a single A100 and handles most production workloads. No API dependency, no per-token costs.”
The Futurist
Big Picture
“Meta is subsidizing open-source AI infrastructure for the entire industry. The strategic value of an open ecosystem outweighs the model training costs.”
The Skeptic
Reality Check
“Benchmarks matching GPT-4 doesn't mean matching GPT-4 in practice. Real-world performance varies significantly by task. Test before committing.”