Model test results

Compare AI models

Checked daily

Compare results, explore different strengths, and discover models beyond the familiar names.

LiveBench tests models on tasks with verifiable answers. Higher is better, on a scale from 0 to 100.

Last checked: Oct 22, 2026Test suite version: 2026-10-1449 models

1Claude Fable 5.1Anthropic87.986

claude-fable-5.1

Long agentic work: 80.800 / 100
Reasoning: 91.700 / 100
Coding: 97.700 / 100
Browser use: 100.000 / 100
Data analysis: 80.300 / 100
Research: 82.000 / 100
Design: 83.400 / 100

Individual tasks

long_agentic: 80.800
reasoning: 91.700
coding: 97.700
browser: 100.000
data_analysis: 80.300
research: 82.000
design: 83.400

Source: LiveBench ↗

2Claude Opus 5.5Anthropic86.529

claude-opus-5.5

Long agentic work: 85.000 / 100
Reasoning: 92.200 / 100
Coding: 96.300 / 100
Browser use: 80.000 / 100
Data analysis: 80.300 / 100
Research: 85.000 / 100
Design: 86.900 / 100

Individual tasks

long_agentic: 85.000
reasoning: 92.200
coding: 96.300
browser: 80.000
data_analysis: 80.300
research: 85.000
design: 86.900

Source: LiveBench ↗

3GPT-6 AstraOpenAI86.143

gpt-6-astra

Long agentic work: 82.400 / 100
Reasoning: 92.700 / 100
Coding: 93.700 / 100
Browser use: 84.000 / 100
Data analysis: 83.000 / 100
Research: 85.200 / 100
Design: 82.000 / 100

Individual tasks

long_agentic: 82.400
reasoning: 92.700
coding: 93.700
browser: 84.000
data_analysis: 83.000
research: 85.200
design: 82.000

Source: LiveBench ↗

4GPT-6.1 SolOpenAI85.414

gpt-6.1-sol

Long agentic work: 73.000 / 100
Reasoning: 92.600 / 100
Coding: 96.000 / 100
Browser use: 88.000 / 100
Data analysis: 82.700 / 100
Research: 83.500 / 100
Design: 82.100 / 100

Individual tasks

long_agentic: 73.000
reasoning: 92.600
coding: 96.000
browser: 88.000
data_analysis: 82.700
research: 83.500
design: 82.100

Source: LiveBench ↗

5Gemini 4 ArgonGoogle84.100

gemini-4-argon

Long agentic work: 69.900 / 100
Reasoning: 95.100 / 100
Coding: 89.700 / 100
Browser use: 92.000 / 100
Data analysis: 85.200 / 100
Research: 83.100 / 100
Design: 73.700 / 100

Individual tasks

long_agentic: 69.900
reasoning: 95.100
coding: 89.700
browser: 92.000
data_analysis: 85.200
research: 83.100
design: 73.700

Source: LiveBench ↗

6Claude Fable 5Anthropic83.614

claude-fable-5

Long agentic work: 53.200 / 100
Reasoning: 89.700 / 100
Coding: 98.300 / 100
Browser use: 92.000 / 100
Data analysis: 80.500 / 100
Research: 86.900 / 100
Design: 84.700 / 100

Individual tasks

long_agentic: 53.200
reasoning: 89.700
coding: 98.300
browser: 92.000
data_analysis: 80.500
research: 86.900
design: 84.700

Source: LiveBench ↗

7Claude Sonnet 5.5Anthropic82.457

claude-sonnet-5.5

Long agentic work: 78.800 / 100
Reasoning: 91.600 / 100
Coding: 94.700 / 100
Browser use: 84.000 / 100
Data analysis: 59.500 / 100
Research: 83.600 / 100
Design: 85.000 / 100

Individual tasks

long_agentic: 78.800
reasoning: 91.600
coding: 94.700
browser: 84.000
data_analysis: 59.500
research: 83.600
design: 85.000

Source: LiveBench ↗

8Claude Opus 5Anthropic82.086

claude-opus-5

Long agentic work: 62.100 / 100
Reasoning: 91.200 / 100
Coding: 90.800 / 100
Browser use: 84.000 / 100
Data analysis: 74.600 / 100
Research: 88.900 / 100
Design: 83.000 / 100

Individual tasks

long_agentic: 62.100
reasoning: 91.200
coding: 90.800
browser: 84.000
data_analysis: 74.600
research: 88.900
design: 83.000

Source: LiveBench ↗

9GPT-6 SolOpenAI81.914

gpt-6-sol

Long agentic work: 66.900 / 100
Reasoning: 88.700 / 100
Coding: 93.100 / 100
Browser use: 92.000 / 100
Data analysis: 81.200 / 100
Research: 69.300 / 100
Design: 82.200 / 100

Individual tasks

long_agentic: 66.900
reasoning: 88.700
coding: 93.100
browser: 92.000
data_analysis: 81.200
research: 69.300
design: 82.200

Source: LiveBench ↗

10DeepSeek V4.1 Flash 1017DeepSeek79.200

deepseek-v4.1-flash-1017

Long agentic work: 50.900 / 100
Reasoning: 89.100 / 100
Coding: 96.000 / 100
Browser use: 76.000 / 100
Data analysis: 79.500 / 100
Research: 79.500 / 100
Design: 83.400 / 100

Individual tasks

long_agentic: 50.900
reasoning: 89.100
coding: 96.000
browser: 76.000
data_analysis: 79.500
research: 79.500
design: 83.400

Source: LiveBench ↗

11Muse Wonder 1.0Meta78.100

muse-wonder-1.0

Long agentic work: 53.200 / 100
Reasoning: 90.600 / 100
Coding: 86.200 / 100
Browser use: 72.000 / 100
Data analysis: 81.600 / 100
Research: 88.000 / 100
Design: 75.100 / 100

Individual tasks

long_agentic: 53.200
reasoning: 90.600
coding: 86.200
browser: 72.000
data_analysis: 81.600
research: 88.000
design: 75.100

Source: LiveBench ↗

12DeepSeek V4.1 FlashDeepSeek78.043

deepseek-v4.1-flash

Long agentic work: 40.100 / 100
Reasoning: 86.700 / 100
Coding: 90.200 / 100
Browser use: 84.000 / 100
Data analysis: 79.300 / 100
Research: 86.800 / 100
Design: 79.200 / 100

Individual tasks

long_agentic: 40.100
reasoning: 86.700
coding: 90.200
browser: 84.000
data_analysis: 79.300
research: 86.800
design: 79.200

Source: LiveBench ↗

13Claude Opus 4.8Anthropic76.771

claude-opus-4.8

Long agentic work: 48.300 / 100
Reasoning: 89.200 / 100
Coding: 77.300 / 100
Browser use: 96.000 / 100
Data analysis: 66.000 / 100
Research: 81.800 / 100
Design: 78.800 / 100

Individual tasks

long_agentic: 48.300
reasoning: 89.200
coding: 77.300
browser: 96.000
data_analysis: 66.000
research: 81.800
design: 78.800

Source: LiveBench ↗

14GPT-5.6 SolOpenAI76.257

gpt-5.6-sol

Long agentic work: 48.500 / 100
Reasoning: 91.700 / 100
Coding: 89.700 / 100
Browser use: 88.000 / 100
Data analysis: 79.800 / 100
Research: 54.500 / 100
Design: 81.600 / 100

Individual tasks

long_agentic: 48.500
reasoning: 91.700
coding: 89.700
browser: 88.000
data_analysis: 79.800
research: 54.500
design: 81.600

Source: LiveBench ↗

15Claude Haiku 5.5Anthropic75.643

claude-haiku-5.5

Long agentic work: 57.600 / 100
Reasoning: 82.400 / 100
Coding: 91.400 / 100
Browser use: 80.000 / 100
Data analysis: 53.100 / 100
Research: 81.900 / 100
Design: 83.100 / 100

Individual tasks

long_agentic: 57.600
reasoning: 82.400
coding: 91.400
browser: 80.000
data_analysis: 53.100
research: 81.900
design: 83.100

Source: LiveBench ↗

16GPT-5.6 TerraOpenAI74.986

gpt-5.6-terra

Long agentic work: 48.400 / 100
Reasoning: 90.600 / 100
Coding: 86.200 / 100
Browser use: 84.000 / 100
Data analysis: 79.300 / 100
Research: 60.400 / 100
Design: 76.000 / 100

Individual tasks

long_agentic: 48.400
reasoning: 90.600
coding: 86.200
browser: 84.000
data_analysis: 79.300
research: 60.400
design: 76.000

Source: LiveBench ↗

17Grok 4.6 xHighxAI74.943

grok-4.6

Long agentic work: 40.700 / 100
Reasoning: 91.200 / 100
Coding: 89.100 / 100
Browser use: 64.000 / 100
Data analysis: 75.300 / 100
Research: 84.500 / 100
Design: 79.800 / 100

Individual tasks

long_agentic: 40.700
reasoning: 91.200
coding: 89.100
browser: 64.000
data_analysis: 75.300
research: 84.500
design: 79.800

Source: LiveBench ↗

18MiMo V2.6 ProXiaomi74.871

mimo-v2.6-pro

Long agentic work: 47.500 / 100
Reasoning: 79.600 / 100
Coding: 81.500 / 100
Browser use: 88.000 / 100
Data analysis: 72.300 / 100
Research: 76.700 / 100
Design: 78.500 / 100

Individual tasks

long_agentic: 47.500
reasoning: 79.600
coding: 81.500
browser: 88.000
data_analysis: 72.300
research: 76.700
design: 78.500

Source: LiveBench ↗

19Muse Spark 1.3Meta74.757

muse-spark-1.3

Long agentic work: 40.400 / 100
Reasoning: 89.700 / 100
Coding: 90.200 / 100
Browser use: 80.000 / 100
Data analysis: 79.600 / 100
Research: 68.600 / 100
Design: 74.800 / 100

Individual tasks

long_agentic: 40.400
reasoning: 89.700
coding: 90.200
browser: 80.000
data_analysis: 79.600
research: 68.600
design: 74.800

Source: LiveBench ↗

20Grok 4.7xAI74.086

grok-4.7

Long agentic work: 50.100 / 100
Reasoning: 82.700 / 100
Coding: 84.500 / 100
Browser use: 60.000 / 100
Data analysis: 76.900 / 100
Research: 86.400 / 100
Design: 78.000 / 100

Individual tasks

long_agentic: 50.100
reasoning: 82.700
coding: 84.500
browser: 60.000
data_analysis: 76.900
research: 86.400
design: 78.000

Source: LiveBench ↗

Ranked by unrounded results. Similar scores do not establish a meaningful difference.

Each exact model setting stays separate. A category mean requires every task in that category; the overall mean requires every category. Tasks have equal weight within a category and categories have equal weight overall. Missing results are never zero. Placements use unrounded values in the chosen column, with shared ties; search retains the placement in the full test. Old scores cannot determine placement. Displayed decimals do not establish a significant difference. This view covers one test suite, not every current model or every kind of ability.

How the comparison works

The overall score shown here is the equal-weight average of all LiveBench categories in the same benchmark release. A total requires a result for every task. Arena ratings stay separate because they measure preference. We do not match similarly named models across sources: versions and reasoning settings can differ. These tests do not cover every model or every real-world task.

Source data: LiveBench; Arena / LMArena (CC BY 4.0). Oossa computes the displayed task averages and presents the results in a simpler format. Arena ↗ · Source ↗ · CC BY 4.0 ↗