Back to glossary

Term

Llama 4 Maverick

Llama 4 Maverick (April 2025) is Meta's large Mixture-of-Experts variant in the Llama 4 lineup — 17B active out of 400B parameters across 128 experts, 1M-token context.

Llama 4 Maverick — explained in more detail

Llama 4 Maverick was released on April 5, 2025 as the large member of the Llama 4 family — alongside Llama 4 Scout. Maverick uses a Mixture-of-Experts architecture with 400 billion total parameters distributed across 128 experts, of which only 17 billion are active per token. MoE and dense layers alternate — experts apply to only half the layers.

Specifications

  • Context window: 1 million tokens.
  • Multimodal: text and image input, early-fusion architecture.
  • License: Llama 4 Community License.
  • Model name: meta-llama/Llama-4-Maverick-17B-128E.

Context

According to Meta, Maverick beats GPT-4o and Gemini 2.0 Flash “across the board” on many multimodal benchmarks and matches the much larger DeepSeek-V3 on reasoning and coding tasks. This positions Maverick as the open alternative in the mid-to-high tier — more capable than Scout but more expensive to run due to the larger total parameter count. Competitors: DeepSeek V4, Mistral Large 3, Qwen 3.5.

Entdecke mehr

Themenuebersicht