Glimmer Index
Plus $4/mo

Claude Sonnet 4

Anthropic · Released May 22, 2025 · Closed weights

Balanced Claude model for coding, analysis, agent workflows, and cost control

Strongest at
Math
Top 44% of the past year's models · one benchmark so far
Context window
200K
tokens in · 64K out
Since the last one
92 days
after Claude Sonnet 3.7
Lab announcement ↗Open in the timeline

Capability profile

Scored against 91 models released in the past year. 50 is the average model; further out is better.

This modelClaude Sonnet 3.7Median of the past year
5090Median of the past yearCoding: 48 · Better than 30%Writing: 54 · Better than 43%Reasoning: 44 · Better than 15%Math: 57 · Top 44%Science: 49 · Better than 7%Claude Sonnet 3.7: 53CodingWritingReasoningMathScience
Coding48Better than 30%
Writing54+1.2Better than 43%
Reasoning44Better than 15%
Math57Top 44% · 1 test
Science49Better than 7%

No results yet: Reading & documents, Visual reasoning, Computer use

Where these scores come from

Each capability combines the published benchmarks listed here, rescaled so 50 is the reference average and 15 points is one standard deviation. A hollow dot rests on a single benchmark.

Since Claude Sonnet 3.7

6 releases in the Claude Sonnet line.

Claude Sonnet 3.7Feb 19, 202592dClaude Sonnet 4May 22, 2025130dClaude Sonnet 4.5Sep 29, 2025141dClaude Sonnet 4.6Feb 17, 2026133dClaude Sonnet 5Jun 30, 202690dClaude Sonnet 5.5Sep 28, 2026
Released92 days later
Context200Ksame
Writing54+1.2

A newer release is out: Claude Sonnet 5.5.

ECI STANDING#113 of 270
SCORE141.56139.0–143.07 CI
FRONTIER LAGLaunched 5.1 months behindScores about what o1 did in December 2024.
SCORES LIKE THESE MODELS
Gemma 4 26B A4BGoogle DeepMind · Apr 2026141.88
o1OpenAI · Dec 2024141.72
Gemini 2.5 Flash (May 2025)Google DeepMind · May 2025141.43
Mistral Medium 3.5Mistral AI · Apr 2026141.36
Full Epoch Capabilities Index ranking ↗

Epoch benchmark boards

Best result on each board it has run. The tick marks the leader.

The Agent Company
33%#3 of 14
MATH level 5
84%#25 of 98
Cybench
35%#9 of 22
GPQA diamond
79%#97 of 234
METR Time Horizons
62%#18 of 42
WeirdML
46%#63 of 141
DTBench
77%#93 of 179
DeepResearch Bench
47%#14 of 26
7 more boards
ARC-AGI-2
6%#63 of 103
SimpleBench
46%#56 of 89
ARC-AGI
40%#70 of 107
GSO-Bench
5%#25 of 35
Fiction.LiveBench
47%#46 of 58
VPCT
34%#26 of 29
GeoBench
37%#26 of 28

Anthropic's past year

12 release days in the past year, about one every 30 days. The next is expected around Nov 12, 2026. Forecasts ↗

NovDecJanFebMarAprMayJunJulAugSepOctNov

Specifications

RELEASE DATEMay 22, 2025
WEIGHTSClosed weights
CONTEXT WINDOW200K
MAX OUTPUT64K
REASONINGNot listed
TOOL CALLINGNot listed
INPUTStext, image/jpeg, image/png, image/gif, image/webp, application/pdf
OUTPUTStext

Models.dev record ↗

Specifications and prices from Models.dev; availability and limits may differ by provider. Benchmarks from Epoch AI (CC BY 4.0), LiveBench and the other sources named above.