Skip to content
Leaderboard / Anthropic

Claude Sonnet 4.5

Anthropic High, 32K, 2025-09-29 CLOSED Proprietary
Other settings:2025-09-29 · 54
Verdict

Claude Sonnet 4.5 is a solid mid-pack model: #115 of 406 by SIscope score. Strongest at following instructions, weaker at math. On the pricey side: costs more than 86% of priced models.

SIscope score
65
#115 of 406
Human-vote rank
#90
of 419 · score 1435
Price per 1M tokens, blended
$6
$3 in · $15 out
Context window
1M
1,000,000 tokens

Shape

Each spoke is a category, 0 at the centre, 100 at the edge.

CodingAgentsno data yetReasoningMathChatInstructions

Category scores

Percentiles: 90 means better than about 90% of tested models.

Coding61
Agents & toolsNo data yet–
Reasoning & knowledge73
Math56
Chat quality79
Following instructions90

What goes into each category · Compare with DeepSeek V4.1 Flash

Human-vote score over time

Text arena, recent updates.

143014351440Oct 20: 1434.5 (rank #89)Oct 22: 1434.5 (rank #89)Oct 28: 1434.5 (rank #90)Oct 20Oct 28

Coding votes over time

WebDev arena, recent updates.

138513931400Oct 16: 1393.7 (rank #97)Oct 18: 1393.8 (rank #98)Oct 19: 1393.7 (rank #101)Oct 22: 1393.9 (rank #102)Oct 23: 1393.8 (rank #103)Oct 27: 1393.8 (rank #104)Oct 28: 1393.5 (rank #105)Oct 16Oct 28

Independent test runs

Run by Epoch AI on the same harness for every model. Bars show the share of questions answered correctly.

GPQA Diamond82%
PhD-level science multiple choice questions
Better than 60% of 333 tested models. Run Oct 21, 2025.
OTIS Mock AIME 2024-2578%
Competition maths problems in the style of AIME
Better than 56% of 313 tested models. Run Oct 21, 2025.
FrontierMath Tiers 1-324%
Very hard, unpublished research-style maths problems
Better than 29% of 122 tested models. Run Jun 11, 2026.

Human-vote arenas

People compare two anonymous answers and pick the better one. Scores work like chess ratings.

ArenaRankScoreRange (95%)VotesUpdated
Text arena
Everyday chat, blind human votes
#90 of 41914351432–143783,892Oct 28, 2026
WebDev arena
Building web apps, blind human votes
#105 of 15213941386–140113,213Oct 28, 2026
Vision arena
Questions about images, blind human votes
Not on this board yet
Agent arena
Real multi-step agent sessions
Not on this board yet

Source: Arena (formerly LMArena) leaderboard dataset, CC BY 4.0. Fetched Nov 4, 2026.

Price

Input
$3
per 1M tokens
Output
$15
per 1M tokens
Blended
$6
3 in : 1 out, per 1M
Context window
1M
max tokens per request
Speed
Coming soon
our own tests

List price from the OpenRouter models API (anthropic/claude-sonnet-4.5), Nov 4, 2026. Providers can charge differently.