Meta Llama 4 vs ChatGPT: Is there a winner?

Meta has released its new Llama 4 language models, entering directly into competition with OpenAI’s ChatGPT. With two of the three Llama 4 variants—Scout and Maverick—already available to try on WhatsApp, Messenger and Instagram, Meta is clearly pushing for mainstream AI adoption. But how do these models stack up against the established leader?

A new herd with different strengths

Meta’s Llama 4 comes in three versions: Scout (lightweight with 109B parameters), Maverick (400B), and Behemoth, which is still in training. While Scout and Maverick are downloadable and already integrated into Meta’s platforms, Behemoth is being optimized for education and STEM benchmarks, with Meta claiming it outperforms GPT-4.5 and Gemini 2.0 Pro on tests like MATH-500 and GPQA Diamond.

Multimodal, but missing features

Both Scout and Maverick support multimodal input, but Meta’s AI currently lacks essential features that ChatGPT and Gemini offer. For instance, there’s no way to upload an image directly—users can only provide URLs. Features like AI search, deep reasoning or advanced image generation are also noticeably absent.

Open, yet controversial

Developers can download Llama 4 from Meta’s site or Hugging Face, a move that reinforces Meta’s commitment to open source. However, the models are still involved in ongoing copyright disputes, with concerns around the use of datasets like LibGen during training. This legal uncertainty casts a shadow over what is otherwise a technically ambitious release.

In short, while Meta Llama 4 shows promise and brings openness and scalability to the AI space, it still trails behind ChatGPT in usability, features, and trust.