Meta has released the latest version of its language model Llama 3. The model is said to perform better than GPT-4o and Anthropic’s Claude 3.5 Sonnet, but does not offer Dutch language support.

Meta releases its new open source AI model Llama 3.1. The latest version of the Meta language model has up to 405 billion parameters and can process up to 128,000 tokens, but smaller variants are also possible. According to the benchmarks, the model should perform better than its competitors GPT-4o and Claude 3.5 Sonnet. The model is available in eight languages, of which Dutch is unfortunately missing.

Better performance

Llama 3.1 is the largest open-source AI model with a maximum of 405 billion parameters and is trained on more than 16,000 Nvidia H100 GPUs. Meta has not yet announced details on the cost of this development, but knowing the price range of the Nvidia chips, you are on the right track.

Mark Zuckerberg shows through some benchmarks that the Meta model outperforms Llama 3.1 GPT-4o and Claude 3.5 Sonnet. Llama 3.1 can handle up to 128,000 tokens, while its predecessor Llama 3, launched by Meta in April, can handle around 8,000 tokens. Llama 3.1 has a maximum of 405 billion parameters, which is new, but Meta is also launching improved versions of the 8B and 70B models.

No Dutch language support

The long-awaited Llama 3.1 sounds promising, but unfortunately does not offer support in Dutch. Dutch-speaking users have to make do with eight other languages ​​that Llama 3.1 supports, according to a page by Hugging Face: English, French, German, Spanish, Italian, Portuguese, Thai and Hindi.

Although Meta strongly emphasizes the open nature of the open-source AI model Llama 3.1, Meta announced last week that the latest version of Llama 3 will not come to Europe. The company does not agree with European legislation on AI. However, the company does not provide any information about the launch of Llama 3.1 in Europe. The model is free to download, but it is not clear in what capacities it will be available in Europe.