Atlas

Models

← All models

Opus 4

Anthropic · 2025-05-22

Opus 4 is Anthropic's Claude 4 flagship, a hybrid-reasoning model built for complex, long-running agentic tasks, research, and coding. It supports extended thinking with tool use and was positioned as a top-tier coding model at launch.

Benchmark scores

BenchmarkScore
A-Fantasia40.3
A-Fantasia (Valid Attempts with Retries)40.0
A-Fantasia Backwards Spelling22.0
A-Fantasia Chess29.0
A-Fantasia Chess (Valid Attempts with Retries)29.0
A-Fantasia Cube (Valid Attempts with Retries)69.1
A-Fantasia Cube Rotation70.0
A-Fantasia Spelling (Valid Attempts with Retries)22.0
AA-LCR36.0
AA-LCR v1.169.3
Active Probing Benchmark (APB)27.6
Active Probing Benchmark (Biology)27.5
Active Probing Benchmark (Chemistry)29.2
Active Probing Benchmark (Computer Science)19.2
Active Probing Benchmark (Economics)31.7
Active Probing Benchmark (Human Disease)40.8
Active Probing Benchmark (Mathematics)18.3
Active Probing Benchmark (Philosophy)27.5
Active Probing Benchmark (Physics)26.7
Aider Polyglot72.0
AIME 2024 (avg@8)48.8
AIME 2024-202541.3
AIME 202575.5
AIME 2025 (Artificial Analysis avg@10)73.3
AIME 2025 (avg@8)33.8
AIME 2025 (MathArena repeated-run mean)70.0
AlgoTune1.3
ARC-AGI-1 Semi-Private35.7
ARC-AGI-28.6
Arena Text — Business, Management & Financial Ops (No Style Control)1337
Loading Atlas data…