Atlas

Models

← All models

Opus 3

Anthropic · 2024-03-04

Claude 3 Opus is Anthropic's most capable model in the Claude 3 family, a multimodal model with text and image input and text output. At launch, Anthropic offered a 200K-token production context, listed 1M-token access for specific use cases, and reported that the Claude 3 models could accept inputs exceeding 1 million tokens.

Benchmark scores

BenchmarkScore
A-Fantasia50.3
A-Fantasia Backwards Spelling35.0
A-Fantasia Chess42.0
A-Fantasia Cube Rotation74.0
Adversarial Robustness13.0
Agentic Tool Use (Chat)48.5
Agentic Tool Use (Enterprise)52.8
AidanBench878
AlpacaEval 2.0 (length-controlled win rate)40.5
AlpacaEval 2.0 (raw win rate)29.1
ARC-Challenge96.4
Arena Text — Business, Management & Financial Ops (No Style Control)1222
Arena Text — Chinese (No Style Control)1247
Arena Text — Creative Writing (No Style Control)1235
Arena Text — English (No Style Control)1263
Arena Text — Entertainment, Sports & Media (No Style Control)1221
Arena Text — Exclude Ties (No Style Control)1175
Arena Text — Exclude Ties (Style Controlled)1262
Arena Text — Expert (No Style Control)1222
Arena Text — French (No Style Control)1274
Arena Text — French (Style Controlled)1338
Arena Text — German (No Style Control)1258
Arena Text — Hard Prompts (No Style Control)1246
Arena Text — Hard Prompts English (No Style Control)1245
Arena Text — Instruction Following (No Style Control)1249
Arena Text — Japanese (No Style Control)1204
Arena Text — Korean (No Style Control)1187
Arena Text — Legal & Government (No Style Control)1264
Arena Text — Life, Physical & Social Science (No Style Control)1254
Arena Text — Longer Query (No Style Control)1259
Loading Atlas data…