Meta Expands Llama AI Model Access for US Allies Across Europe and Asia
Meta is preparing to make its advanced
Llama large language model broadly accessible to US allies in Europe and Asia by 2025, marking a significant expansion of its artificial intelligence strategy. This move is set to deepen international cooperation around cutting-edge AI technology developed by Meta, the parent company of Facebook and Instagram[2][4].
Llama 4: The Latest Evolution
The latest version,
Llama 4, was released in April 2025 and brings a new architecture based on a mixture of experts, capable of both text and image input, and supporting 12 languages. Two main versions were launched:
- Scout: 17 billion active parameters, 16 experts, 10M context window, with 109B total parameters.
- Maverick: 17 billion active parameters, 128 experts, 1M context window, with 400B total parameters.
A third, unreleased version called Behemoth is reportedly in ongoing development with 288 billion active parameters and nearly 2 trillion overall[2][1].
Key Features and Training Approach
- Multimodal and Multilingual: Llama 4 processes both text and images and handles 12 languages natively for wider accessibility.
- Instruction-Tuned Models: Available as both base and instruction-tuned versions for improved adaptability and usability.
- Training Data: Meta uses a combination of publicly available datasets, licensed data, and proprietary sources like public Facebook and Instagram posts to train Llama 4. Its knowledge was last updated in August 2024[2].
Benchmark Results and Controversies
Meta claims that Llama 4 outperformed
ChatGPT (specifically GPT-4o) on the LMArena AI benchmark. However, this was achieved using an unpublished, "experimental chat version" of Llama 4 optimized for conversational interaction. LMArena clarified that its policies would be updated to prevent such outlier benchmarking in the future, and some critics on social media accused Meta of using test data to enhance benchmarks, a claim the company denies[2].
Global Impact and Future Plans
Currently, Meta’s AI products, including
Meta AI, reach hundreds of millions of users monthly. With the planned rollout of Llama 4 in Europe and Asia, Meta anticipates further growth in AI adoption. The company also aims to solidify its position against competitors like
ChatGPT by fostering an open ecosystem for language models and AI assistants[1][3].
Conclusion
Meta’s expanded Llama 4 global availability is a pivotal step in its strategy to power AI innovation and international collaboration. By providing advanced AI infrastructure and tools, Meta is positioning itself as a key architect in the global AI landscape, empowering partners across major regions while contending with ongoing industry scrutiny.
