Meta AI (Llama)

Multimodal model

Meta's assistant and the open-weight Llama models that power much of the local/open-source ecosystem.

Version timeline as of Jun 25, 2026.

Version timeline

11 versions

Llama 4 MaverickCurrentDefault
Released Apr 1, 2025Llama-4-Maverick-17B-128E-Instruct-FP8· 1M ctx · cutoff Apr 1, 2025 · Flagship
Flagship open-weight MoE model: 17B active params, 128 experts, 400B total. Natively multimodal; strongest general-purpose Llama. Llama 4 Community License.
Llama 4 ScoutActive
Released Apr 1, 2025· 10M ctx · cutoff Apr 1, 2025 · Open Weight
MoE: 17B active params, 16 experts, 109B total. 10M-token context (longest of any released model); best for needle-in-haystack retrieval. Runs on a single H100.