Atlas

Models

← All models

Llama 3.1 405B Instruct

Meta · 2024-07-23 · 405.9B parameters

Llama 3.1 405B Instruct is the instruction-tuned version of Meta's 405-billion-parameter Llama 3.1 model, released in July 2024. Fine-tuned with supervised fine-tuning and reinforcement learning from human feedback, it is optimized for general-purpose chat, reasoning, multilingual tasks, and tool use with a 128K-token context window.

Benchmark scores

BenchmarkScore
A-Fantasia68.7
A-Fantasia Backwards Spelling100.0
A-Fantasia Chess38.0
A-Fantasia Cube Rotation68.0
AA-LCR24.3
AA-Omniscience Index-17.3
Adversarial Robustness10.0
Agentic Tool Use (Chat)40.1
Agentic Tool Use (Enterprise)50.4
AidanBench759
AIME 20253.0
Arabic875
Arena Text — Business, Management & Financial Ops (No Style Control)1245
Arena Text — Chinese (No Style Control)1241
Arena Text — Creative Writing (No Style Control)1260
Arena Text — English (No Style Control)1312
Arena Text — Entertainment, Sports & Media (No Style Control)1242
Arena Text — Exclude Ties (No Style Control)1214
Arena Text — Exclude Ties (Style Controlled)1286
Arena Text — Expert (No Style Control)1228
Arena Text — French (No Style Control)1278
Arena Text — French (Style Controlled)1340
Arena Text — German (No Style Control)1253
Arena Text — Hard Prompts (No Style Control)1270
Arena Text — Hard Prompts English (No Style Control)1293
Arena Text — Instruction Following (No Style Control)1259
Arena Text — Japanese (No Style Control)1209
Arena Text — Korean (No Style Control)1173
Arena Text — Legal & Government (No Style Control)1302
Arena Text — Life, Physical & Social Science (No Style Control)1295
Loading Atlas data…