Meta

4 active models · 1 deprecated · Official website →

Active Models

Llama 4 Maverick

✅ Active

400B MoE model with 17 active experts, 1M native context

Context: 1M

Input: Free/1M

Output: Free/1M

textvisioncodemultilingual

Llama 4 Scout

✅ Active

109B MoE model with 10M context window, 16 active experts

Context: 10M

Input: Free/1M

Output: Free/1M

textvisioncodemultilingual

Llama 3.3

✅ Active

70B model matching Llama 3.1 405B performance at fraction of compute

Llama 3.1

✅ Active

First open-source model to rival GPT-4 class, available in 8B/70B/405B

Context: 128K

Input: Free/1M

Output: Free/1M

textcodemultilingualfunction-calling

Deprecated Models

Llama 2

⚠️ Deprecated

Legacy open-source model, replaced by Llama 3 series

Change History

Apr 5, 2025New ReleaseMeta

Llama 4 Maverick

Llama 4 Maverick released as open-source 400B MoE model

Apr 5, 2025New ReleaseMeta

Llama 4 Scout

Llama 4 Scout launched with industry-leading 10M token context

Dec 6, 2024New ReleaseMeta

Llama 3.3

Llama 3.3 70B released, matching 405B quality at lower cost

Dec 1, 2024DeprecationMeta

Llama 2

Deprecated — migrate to Llama 3.x series

Sep 1, 2024UpdateMeta

Llama 3.1

Quantized and distilled variants released by community

Jul 23, 2024New ReleaseMeta

Llama 3.1

Llama 3.1 released in 8B, 70B, and 405B variants

Jul 18, 2023New ReleaseMeta

Llama 2

Llama 2 released in 7B, 13B, and 70B variants