Atlas

Models

← All models

Claude 3 Sonnet

Anthropic · 2024-03-04

Claude 3 Sonnet is Anthropic's mid-tier multimodal model in the Claude 3 family, balancing intelligence and speed for enterprise workloads. It accepts text and image inputs and produces text output. Anthropic offered a 200K-token production context at launch while reporting that all Claude 3 models could accept inputs exceeding 1 million tokens.

Benchmark scores

BenchmarkScore
Agentic Tool Use (Chat)40.4
Agentic Tool Use (Enterprise)54.2
AidanBench548
AlpacaEval 2.0 (length-controlled win rate)34.9
AlpacaEval 2.0 (raw win rate)25.6
ARC-Challenge93.2
Arena Text — Business, Management & Financial Ops (No Style Control)1174
Arena Text — Chinese (No Style Control)1188
Arena Text — Creative Writing (No Style Control)1187
Arena Text — English (No Style Control)1226
Arena Text — Entertainment, Sports & Media (No Style Control)1172
Arena Text — Exclude Ties (No Style Control)1109
Arena Text — Exclude Ties (Style Controlled)1199
Arena Text — Expert (No Style Control)1172
Arena Text — French (No Style Control)1228
Arena Text — French (Style Controlled)1301
Arena Text — German (No Style Control)1204
Arena Text — Hard Prompts (No Style Control)1197
Arena Text — Hard Prompts English (No Style Control)1204
Arena Text — Instruction Following (No Style Control)1200
Arena Text — Japanese (No Style Control)1132
Arena Text — Korean (No Style Control)1128
Arena Text — Legal & Government (No Style Control)1214
Arena Text — Life, Physical & Social Science (No Style Control)1210
Arena Text — Longer Query (No Style Control)1211
Arena Text — Medicine & Healthcare (No Style Control)1187
Arena Text — Multi-turn (No Style Control)1227
Arena Text — Non-English (No Style Control)1205
Arena Text — Non-English (Style Controlled)1267
Arena Text — Occupational Mathematical (No Style Control)1213
Loading Atlas data…