Atlas

Models

← All models

Llama 4 Scout Instruct

Meta · 2025-04-05 · 108.6B parameters

Llama 4 Scout Instruct is Meta's instruction-tuned, natively multimodal mixture-of-experts checkpoint with 16 routed experts plus a shared expert and native text-and-image input. Meta describes Scout as 17B activated and 109B total parameters (both rounded), with roughly 40T pretraining tokens and an August 2024 data cutoff; the official artifact configuration supports up to 10,485,760 tokens.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo0.0
AA-Briefcase v1.1 Elo0.0
AA-LCR25.8
AA-LCR v1.127.7
AA-Omniscience Index-52.1
AidanBench334
AIME 2024 (avg@8)27.9
AIME 2024-202519.0
AIME 2025 (Artificial Analysis avg@10)14.0
AIME 2025 (avg@8)10.0
ARC-AGI-1 Semi-Private0.5
ARC-AGI-20.0
Arena Text — Business, Management & Financial Ops (No Style Control)1257
Arena Text — Chinese (No Style Control)1255
Arena Text — Creative Writing (No Style Control)1249
Arena Text — English (No Style Control)1304
Arena Text — Entertainment, Sports & Media (No Style Control)1242
Arena Text — Exclude Ties (No Style Control)1208
Arena Text — Exclude Ties (Style Controlled)1268
Arena Text — Expert (No Style Control)1235
Arena Text — French (No Style Control)1283
Arena Text — French (Style Controlled)1325
Arena Text — German (No Style Control)1272
Arena Text — Hard Prompts (No Style Control)1266
Arena Text — Hard Prompts English (No Style Control)1293
Arena Text — Instruction Following (No Style Control)1248
Arena Text — Japanese (No Style Control)1206
Arena Text — Korean (No Style Control)1207
Arena Text — Legal & Government (No Style Control)1285
Arena Text — Life, Physical & Social Science (No Style Control)1278
Loading Atlas data…