Sign in

Llama-3.1-8B-Instruct

meta-llama

Hugging FaceArtificial Analysis

BenchClub cloud

Pass rate
71%
32 of 45 checks
Speed
124.4 tok/s
decode, short prompt
At 8k context
97.0 tok/s
decode
Cloud runs
5
speeds within 1.2%
  • Coding8/1553%
  • Tool use9/1560%
  • Long context15/15100%
Suite
[email protected]
Machine
NVIDIA L40S · 48 GB
Engine
ollama 0.40.14

Headline numbers come only from cloud runs. Runs on members’ own machines are listed below and don’t change them.

Arena score
1187
Among open models
#110of 144
Among all on Arena
#330
Votes
49,605

Measured on Llama 3.1 8B Instruct, the served version, not a local build. LMArena, CC BY 4.0, published Oct 28, 2026.

Score by category

CategoryScore
  • English1218
  • Legal & government1206
  • Hard prompts, English1196
  • Software & IT1195
  • Coding1195
  • Science1193
  • Medicine & health1182
  • Long prompts1181
  • Applied math1180
  • Math1179
  • French1178
  • Hard prompts1175
  • Multi-turn1172
  • Spanish1169
  • Business & finance1166
  • Writing & literature1162
  • Instructions1159
  • Russian1158
  • Creative writing1154
  • Chinese1151
  • Non-English1148
  • Media & sports1145
  • German1144
  • Expert1144
  • Japanese1061
  • Excluding ties1055
  • Korean1053

BenchClub runs

Run this model
Run by

PassedFailedNot supported