EN简中

Evaluated on private datasetsCompare how models actually perform

Compare capability scores, evaluation time, and reference cost to find the right model and reasoning effort.

View full rankingsCompare modelsHow we evaluate
6,627,737Total visits

15 providers · 44 models · 95 configs

Leading model by provider

Overall
Anthropicclaude-opus-5-589.15OpenAIgpt-6-astra88.80Metamuse-wonder81.55Moonshotk3.1-pro79.60GoogleGemini 4 Argon79.35xAIgrok-4.778.55
All 15 providers
DeepSeekdeepseek-v4.1-flash-101774.30MiniMaxMiniMax-M3.1-Flash73.60Zhipuglm-5.370.40Xiaomi MiMomimo-v2.6-pro69.25Tencent Hunyuanhy4-preview66.70StepFunstep-565.20QwenQwen 3.8 Flash65.15Doubaodoubao-seed-2-1-pro61.55NVIDIANemotron-3-Ultra44.05

Model rankingsOverall

44 models · 95 configs

Full rankings →
Scoring dimension
RankModelCompare
01
claude-opus-5-5
89.1584971007370
02
gpt-6-astra
88.809596957561
03
claude-fable-5-1
87.258591958068
04
gpt-6.1-sol
86.609596906670
05
claude-sonnet-5-5
82.308689857366
06
claude-opus-5
81.707087957361
07
muse-wonder
81.558284907359
08
gpt-6.1-terra
80.058984856461
09
k3.1-pro
79.608491806864
10
Gemini 4 Argon
79.358284857059
  1. 01
    claude-opus-5-5
    89.15Overall
    Axis scores
    Overall
    89.15
    Backend engineering
    84
    Frontend interaction
    97
    Knowledge & Reasoning
    100
    Visual aesthetics
    73
    Maintainability
    70
  2. 02
    gpt-6-astra
    88.80Overall
    Axis scores
    Overall
    88.80
    Backend engineering
    95
    Frontend interaction
    96
    Knowledge & Reasoning
    95
    Visual aesthetics
    75
    Maintainability
    61
  3. 03
    claude-fable-5-1
    87.25Overall
    Axis scores
    Overall
    87.25
    Backend engineering
    85
    Frontend interaction
    91
    Knowledge & Reasoning
    95
    Visual aesthetics
    80
    Maintainability
    68
  4. 04
    gpt-6.1-sol
    86.60Overall
    Axis scores
    Overall
    86.60
    Backend engineering
    95
    Frontend interaction
    96
    Knowledge & Reasoning
    90
    Visual aesthetics
    66
    Maintainability
    70
  5. 05
    claude-sonnet-5-5
    82.30Overall
    Axis scores
    Overall
    82.30
    Backend engineering
    86
    Frontend interaction
    89
    Knowledge & Reasoning
    85
    Visual aesthetics
    73
    Maintainability
    66
  6. 06
    claude-opus-5
    81.70Overall
    Axis scores
    Overall
    81.70
    Backend engineering
    70
    Frontend interaction
    87
    Knowledge & Reasoning
    95
    Visual aesthetics
    73
    Maintainability
    61
  7. 07
    muse-wonder
    81.55Overall
    Axis scores
    Overall
    81.55
    Backend engineering
    82
    Frontend interaction
    84
    Knowledge & Reasoning
    90
    Visual aesthetics
    73
    Maintainability
    59
  8. 08
    gpt-6.1-terra
    80.05Overall
    Axis scores
    Overall
    80.05
    Backend engineering
    89
    Frontend interaction
    84
    Knowledge & Reasoning
    85
    Visual aesthetics
    64
    Maintainability
    61
  9. 09
    k3.1-pro
    79.60Overall
    Axis scores
    Overall
    79.60
    Backend engineering
    84
    Frontend interaction
    91
    Knowledge & Reasoning
    80
    Visual aesthetics
    68
    Maintainability
    64
  10. 10
    Gemini 4 Argon
    79.35Overall
    Axis scores
    Overall
    79.35
    Backend engineering
    82
    Frontend interaction
    84
    Knowledge & Reasoning
    85
    Visual aesthetics
    70
    Maintainability
    59

Performance & consumption

Highest-scoring tested effort per model · Overall

Model origin
Reference cost
405060708090100$0.1$0.3$1$3$10$30Score ↑Reference cost (USD) · log scaleHigher score / lower useHigher score / higher useLower score / lower useLower score / higher usegpt-6.1-sol · 86.6 · $1.69claude-sonnet-5-5 · 82.3 · $2.8gpt-6.1-terra · 80.05 · $0.85gpt-6-sol · 78.9 · $1.1grok-4.6 · 76.85 · $2.51deepseek-v4.1-flash-1017 · 74.3 · $0.45gpt-5.6-sol · 74.15 · $3.25MiniMax-M3.1-Flash · 73.6 · $0.21muse-spark-1.3 · 72.6 · $0.07claude-opus-5-5 · 89.15 · $13.41gpt-6-astra · 88.8 · $7.01claude-fable-5-1 · 87.25 · $26.15claude-opus-5 · 81.7 · $7.99muse-wonder · 81.55 · $6.78k3.1-pro · 79.6 · $5grok-4.7 · 78.55 · $5.25claude-opus-4-8 · 76.8 · $9.18k3 · 71.6 · $3.28glm-5.3 · 70.4 · $1.83mimo-v2.6-pro · 69.25 · $0.66deepseek-v4.1-flash · 67.85 · $0.41hy4-preview · 66.7 · $1.63gpt-5.6-terra · 65.9 · $1.94claude-haiku-5-5 · 65.65 · $0.21step-5 · 65.2 · $1.24Qwen 3.8 Flash · 65.15 · $0.42deepseek-v4-pro · 65.15 · $1.15k2.8 · 63.55 · $0.66gpt-6-luna · 62 · $0.06doubao-seed-2-1-pro · 61.55 · $1.38deepseek-v4-flash · 60.5 · $0.42deepseek-v4-flash-vision-exp · 59.95 · $0.32mimo-v2.6-flash · 56.1 · $0.28Qwen 3.8 27b · 53 · $1.79gpt-5.6-luna · 52.35 · $0.16Gemini 3.7 Flash High · 51.95 · $0.13
Browse chart models · 36
Analysis options & method
Score median 71 · Consumption median $3.27

Score boundary uses models with available reference costs; resource boundary is the median among higher-scoring models. Only complete positive published values are plotted. Comparison method

Full rankings →

Generation & effort comparisons

Overall score
Explore axes and subscores

What the five dimensions measure

Explore the evaluation
Backend engineeringService logic, error handling, and test verification.
Frontend interactionInterface behavior, user interactions, and state feedback.
Knowledge & ReasoningCross-disciplinary knowledge and complex problem solving.
Visual aestheticsComposition, typography, color palette, and control craft in the finished UI.
MaintainabilityCode readability, modular organization, and cost of change.

Evaluated on our private datasets; scores reflect these specific tasks. Why we test this way