Atlas

Models

← All models

Llama 3.1 8B Instruct

Meta · 2024-07-23 · 8B parameters

Llama 3.1 8B Instruct is an 8-billion-parameter instruction-tuned model released by Meta in July 2024 as part of the Llama 3.1 update. It extends the Llama 3 8B Instruct with improved multilingual capabilities, a 128K-token context window, and enhanced performance on reasoning and agentic tasks. Llama 3.1 prompt formats include tool-calling workflows, though Meta recommends the 70B and 405B instruct models for applications combining conversation and tool calls.

Benchmark scores

BenchmarkScore
AA-LCR15.7
AA-Omniscience Index-30.3
Agentic Tool Use (Chat)6.1
Agentic Tool Use (Enterprise)17.4
AidanBench398
AIME 20254.3
Arena Text — Business, Management & Financial Ops (No Style Control)1166
Arena Text — Chinese (No Style Control)1150
Arena Text — Creative Writing (No Style Control)1155
Arena Text — English (No Style Control)1218
Arena Text — Entertainment, Sports & Media (No Style Control)1145
Arena Text — Exclude Ties (No Style Control)1056
Arena Text — Exclude Ties (Style Controlled)1090
Arena Text — Expert (No Style Control)1143
Arena Text — French (No Style Control)1176
Arena Text — French (Style Controlled)1205
Arena Text — German (No Style Control)1144
Arena Text — Hard Prompts (No Style Control)1175
Arena Text — Hard Prompts English (No Style Control)1197
Arena Text — Instruction Following (No Style Control)1159
Arena Text — Japanese (No Style Control)1061
Arena Text — Korean (No Style Control)1053
Arena Text — Legal & Government (No Style Control)1206
Arena Text — Life, Physical & Social Science (No Style Control)1194
Arena Text — Longer Query (No Style Control)1182
Arena Text — Medicine & Healthcare (No Style Control)1181
Arena Text — Multi-turn (No Style Control)1172
Arena Text — Non-English (No Style Control)1148
Arena Text — Non-English (Style Controlled)1178
Arena Text — Occupational Mathematical (No Style Control)1180
Loading Atlas data…