NEWHow AI models draw the world map · Gecko Tests
Beta
Home/Models/Claude Opus 4
Anthropic logo

Claude Opus 4

by Anthropic · Released May 2025

Multimodal
49.8
avg score
Rank #146
Compare
Better than 53% of all models
Context
200K tokens (~100 books)
Input $/1M
$15.00
Output $/1M
$75.00
Type
multimodal
License
Proprietary
Benchmarks
23 tested
Data as of
About

Claude Opus 4 is benchmarked as the world’s best coding model, at time of release, bringing sustained performance on complex, long-running tasks and agent workflows. It sets new benchmarks in...

Tested on 23 benchmarks · BenchGecko score 49.8. Top scores: MATH level 5 (85.0%), Lech Mazur Writing (83.6%), Aider polyglot (72.0%).

Looking for similar performance at lower cost?
Qwen3 Next 80B A3B Instruct scores 49.2 (99% as good) at $0.09/1M input · 99% cheaper
Capabilities
coding
49.8
#92 globally
reasoning
31.6
#118 globally
math
41.1
#152 globally
knowledge
46.3
#154 globally
general
59.1
#16 globally
Benchmark Scores
Compare All
Tested on 23 benchmarks · Ranked across 5 categories
Score Distribution (all 312 models)
0255075100
▲ You are here
codingCompare coding →
Aider polyglot

Multi-language code editing from Aider. Tests editing ability across Python, JavaScript, TypeScript, Java, C++, Go, Rust, and more.

72.0·
SWE-Bench verified

Real-world software engineering tasks from GitHub issues. Models must diagnose bugs and write patches that pass test suites. Human-verified subset of SWE-bench.

70.7·
SWE-Bench Verified (Bash Only)

SWE-bench Verified solved using only bash commands, no specialized frameworks. Tests raw terminal-based problem solving.

67.6·
reasoningCompare reasoning →
SimpleBench

Deceptively simple questions that humans find easy but AI models often get wrong. Tests common sense and reasoning gaps.

50.6·
ARC-AGI

Abstraction and Reasoning Corpus. Tests fluid intelligence through novel visual pattern recognition puzzles. Core measure of general intelligence.

35.7·
ARC-AGI-2

ARC-AGI 2, harder sequel to ARC. More complex abstract reasoning patterns that test generalization ability beyond training data.

8.6·
mathCompare math →
MATH level 5

Competition-level math from AMC, AIME, and olympiad problems. Level 5 is the hardest tier, requiring creative problem-solving.

85.0·
OTIS Mock AIME 2024-2025

Mock AIME (American Invitational Mathematics Exam) problems from OTIS. Tests mathematical competition performance.

64.4·
FrontierMath-2025-02-28-Private

Original research-level math problems created by professional mathematicians. Problems are unpublished and cannot be memorized.

7.9·
Quick compare:Meta logovs Llama 3.1 70B InstructMistral AI logovs Mistral 7B V0.1Google DeepMind logovs Gemma 2 27B
Excellent (85+) Good (70-85) Average (50-70) Below (<50)
Model Family · Anthropic Claude Opus
Jan 2024
28.1
n/an/actx13 benchmarks
May 2025
49.8+21.7
$15.00/M in200Kctx23 benchmarks
Aug 2025
42.3-7.5
$15.00/M in200Kctx23 benchmarks
Aug 2025
$7.50/M in(-7.50)200Kctx
Nov 2025
59.0+16.7
$5.00/M in(-2.50)200Kctx42 benchmarks
Nov 2025
$2.50/M in(-2.50)200Kctx
Feb 2026
66.4+7.4
$5.00/M in(+2.50)1.0Mctx(+800K)40 benchmarks
Feb 2026
$2.50/M in(-2.50)1.0Mctx
Apr 2026
83.3+16.9
$30.00/M in(+27.50)1.0Mctx12 benchmarks
Apr 2026
65.6-17.7
$5.00/M in(-25)1.0Mctx30 benchmarks
Apr 2026
$2.50/M in(-2.50)1.0Mctx
May 2026
$30.00/M in(+27.50)1.0Mctx3 benchmarks
May 2026
71.8+6.2
$5.00/M in(-25)1.0Mctx29 benchmarks
May 2026
$2.50/M in(-2.50)1.0Mctx
May 2026
$10.00/M in(+7.50)1.0Mctx3 benchmarks
Jul 2026
86.5+14.7
$5.00/M in(-5)1.0Mctx23 benchmarks
Jul 2026
$2.50/M in(-2.50)1.0Mctx
Jul 2026
$10.00/M in(+7.50)1.0Mctx
Sep 2026
96.8+10.3
$4.00/M in(-6)1.0Mctx24 benchmarks
Sep 2026
$2.00/M in(-2)1.0Mctx
See the full Claude Opus family →
Similar Models
Meta logo
Llama 3.1 70B Instruct
Meta
49.7$0.40/1M
Mistral AI logo
Mistral 7B V0.1
Mistral AI
50.0n/a
Google DeepMind logo
Gemma 2 27B
Google DeepMind
50.1$0.65/1M
Links
Info
AnthropicPricing explorerDevelopers · API
Research
Technical Report
Documentation
API DocsPlayground
Community
@Anthropic
BenchGecko API
claude-opus-4
Specifications
  • Typemultimodal
  • Context200K tokens (~100 books)
  • ReleasedMay 2025
  • LicenseProprietary
  • StatusActive
  • Cost / Message~$0.105
Available On
Anthropic logoAnthropic$15.00
Categories
codingreasoningmathknowledgegeneral
Learn More
context-windowtransformertokens
Share & Export
Tweet
Related Models
Meta logoLlama 3.1 70B Instruct
49.7
Mistral AI logoMistral 7B V0.1
50.0
Google DeepMind logoGemma 2 27B
50.1
TII logoFalcon-180B
50.3
Claude Opus 4 is a proprietary multimodal AI model by Anthropic, released in May 2025. It has an average benchmark score of 49.8. Context window: 200K tokens.

Related Models

Llama 3.1 70B Instruct · MetaMistral 7B V0.1 · Mistral AIGemma 2 27B · Google DeepMindFalcon-180B · TIIQwen3 Max · Alibaba Qwen

Benchmarks

MATH level 5Lech Mazur WritingAider polyglotSWE-Bench verifiedDtbench

Related Pages

Anthropic · ProviderAnthropic · EconomyAll ModelsCompare ModelsPricingDevelopers · APIContext Window · Glossary

Key facts · as of 2026-09-21

  • Claude Opus 4 by Anthropic. BenchGecko score 49.8, rank 146 of 312 scored models (normalized average of public benchmark scores).
  • List price $15.00 input · $75.00 output per 1M tokens (as of 2026-09-21).

How to cite · data as of 2026-09-21

Claude Opus 4 · benchmarks, pricing and providers. BenchGecko, data as of 2026-09-21. https://benchgecko.ai/model/claude-opus-4

Credit "Source: BenchGecko" with a link. Prices per provider and Gecko Tests are BenchGecko data (CC BY 4.0); benchmark scores keep their original source, listed in the JSON. JSON · llms.txt · MCP