Atlas

Models

← All models

Sonnet 4

Anthropic · 2025-05-22

Sonnet 4 is Anthropic's balanced mid-tier Claude 4 model, a hybrid-reasoning model offering standard inference plus optional extended thinking. The successor to Claude 3.7 Sonnet, it is tuned for strong coding, parallel tool use, and general-purpose work.

Benchmark scores

BenchmarkScore
A-Fantasia44.0
A-Fantasia Backwards Spelling32.0
A-Fantasia Chess29.0
A-Fantasia Cube Rotation71.0
AA-LCR64.7
AA-Omniscience Index0.2
Agent Red Teaming k=100 (CAIS)86.6
Aider Polyglot61.3
AIME 2024 (avg@8)82.5
AIME 2024-202576.3
AIME 202574.3
AIME 2025 (avg@8)70.0
ALE-Bench1088
APEX-Agents9.3
ARC-AGI-140.0
ARC-AGI-22.1
Arena Text — Business, Management & Financial Ops (No Style Control)1317
Arena Text — Chinese (No Style Control)1343
Arena Text — Creative Writing (No Style Control)1339
Arena Text — English (No Style Control)1346
Arena Text — Entertainment, Sports & Media (No Style Control)1320
Arena Text — Exclude Ties (No Style Control)1293
Arena Text — Exclude Ties (Style Controlled)1368
Arena Text — Expert (No Style Control)1334
Arena Text — French (No Style Control)1363
Arena Text — French (Style Controlled)1426
Arena Text — German (No Style Control)1334
Arena Text — Hard Prompts (No Style Control)1354
Arena Text — Hard Prompts English (No Style Control)1363
Arena Text — Instruction Following (No Style Control)1350
Loading Atlas data…