Skip to main content
Dapols

AI benchmarks

AI models, compared on quality and cost.

A practical view of published model scores that helps you compare capability, category strengths, and run cost without losing the detail.

See the value chartBenchmark source Latest selected release · Jun 25, 2026

Which model should I pay for?

Score against cost per task, at every effort setting.

Each line is one model and each dot on it is a reasoning-effort setting, from low to max. Up is smarter, left is cheaper. The blue staircase links the settings that nothing cheaper beats.

How to read this

  1. Up means a higher Intelligence Index score; right means more money per task. The cost axis is logarithmic, so each gridline is ten times the one before.
  2. A dot on the blue staircase is a best buy at its price: nothing cheaper scores higher. Anything below the staircase is beaten by something that costs less.
  3. A task is one of the benchmark’s test jobs, many of them long and multi-step. Use the cost to compare models, not to budget: an everyday email costs a small fraction of it.

Intelligence vs cost per task

Keyboard: tab into the chart, then left and right move along a line, up and down move between lines, Enter selects and Escape clears.

Every plotted point: model, effort setting, Intelligence Index, cost per task, and whether it is a best buy.
ModelEffortIndexCost per taskBest buy
Claude Fable 5.2low54$3.10No
Claude Fable 5.2medium56$3.80Yes
Claude Fable 5.2high62$9.08Yes
Claude Fable 5.2xhigh62$14.83Yes
Claude Fable 5.2max65$26.54Yes
Claude Opus 5.5low35$0.98No
Claude Opus 5.5medium48$2.45No
Claude Opus 5.5high52$3.47No
Claude Opus 5.5xhigh57$6.05Yes
Claude Opus 5.5max59$10.04No
Claude Sonnet 5.5low26$0.51No
Claude Sonnet 5.5medium31$0.78No
Claude Sonnet 5.5high42$1.64No
Claude Sonnet 5.5xhigh50$4.09No
Claude Sonnet 5.5max55$10.02No
Claude Haiku 5.5low19$0.044Yes
Claude Haiku 5.5medium25$0.092No
Claude Haiku 5.5high30$0.16No
Claude Haiku 5.5xhigh35$0.25No
Claude Haiku 5.5max37$0.41No
GPT-6 Astralow43$1.42No
GPT-6 Astramedium52$2.73No
GPT-6 Astrahigh55$3.24Yes
GPT-6 Astraxhigh55$3.66Yes
GPT-6 Astramax57$6.25No
GPT-6.1 Sollow35$0.16No
GPT-6.1 Solmedium46$0.36Yes
GPT-6.1 Solhigh50$0.59Yes
GPT-6.1 Solxhigh53$0.76Yes
GPT-6.1 Solmax55$1.27Yes
GPT-6.1 Terralow29$0.071Yes
GPT-6.1 Terramedium38$0.17Yes
GPT-6.1 Terrahigh42$0.26Yes
GPT-6.1 Terraxhigh45$0.40No
GPT-6.1 Terramax51$0.96No
GPT-6 Lunalow10$0.005Yes
GPT-6 Lunamedium18$0.042Yes
GPT-6 Lunahigh21$0.073No
GPT-6 Lunaxhigh23$0.10No
GPT-6 Lunamax28$0.15No
Gemini 4 Carbonlow54$7.19No
Gemini 4 Carbonmedium58$6.59Yes
Gemini 4 Carbonhigh60$7.76Yes
Gemini 4 Argonhigh50$9.95No
Gemini 3.8 Flashmedium30$6.26No
Gemini 3.8 Flashhigh31$7.77No
DeepSeek V4.1 Flash 1017high29$0.46No
DeepSeek V4.1 Flash 1017max33$0.60No
Grok 4.7low31$2.92No
Grok 4.7high35$6.18No
Grok 4.7xhigh36$7.77No
Kimi K3.1 Prodefault47$3.68No
Muse Ember 1.0max36$0.14Yes
Mistral Large 4medium23$1.41No
Mistral Large 4high26$2.06No

Sources and notes

Every dot is copied from its own Artificial Analysis model page: the Intelligence Index and Artificial Analysis’s published cost per index task, at the vendor’s list price with caching included. Dapols computes only the staircase, the best-value pick and the comparison advice. Only the effort settings Artificial Analysis tested are drawn, so a short line means fewer published settings, not a missing measurement.

Gemini 4 Carbon’s line doubles back: Artificial Analysis measured its low setting costing more per task than medium. It is drawn as published.

Gemini 3.8 Flash at low effort scores 23 on Artificial Analysis, which has not published its cost per task, so it is not plotted. artificialanalysis.ai

How Artificial Analysis runs the index

Latest benchmark & model updates

The leaderboard on this page is LiveBench's 25 June release, re-captured on 22 September 2026. It now includes Grok 4.7 xHigh and DeepSeek V4.1 Flash Max Effort. The model notes below are from the vendors' own pages, labelled as such.

Grok 4.6

Same $2 / $6 API price as Grok 4.5. xAI reports an Artificial Analysis Intelligence Index of 61 — tied with GPT-5.6 Sol Max, one point behind Fable 5 Max. Strong on knowledge-work evals; 26% on Terminal-Bench v3.

Read the Grok 4.6 briefing

DeepSeek V4.1 Flash

DeepSeek V4.1 Flash replaced V4 Flash on 10 Sep 2026 on the new deepseek-flash API id, at $0.15 / $0.60 per 1M tokens off-peak and $0.30 / $1.20 at peak. From 14 Sep, V4 Pro requests are routed to it too, so switch model ids now. LiveBench now scores V4.1 Flash Max Effort at 81.1 overall.

See it in the model tracker

Leaderboard rows come only from published LiveBench data. We do not invent scores.

Best overall

Claude Fable 5.1 Max Effort leads this snapshot at 83.4 overall, $1.212 per successful task.

Best value per task

ox-alpha-max runs at $0.000 per successful task and still scores 69.2 overall.

Best open weights

DeepSeek V4.1 Flash Max Effort is the strongest open-weight model here at 81.1 overall — the option you can host yourself.

Best per capability

If your work is mostly one kind of task, these lead their category in this snapshot.

Reasoning
GPT-6 Astra Max Effort 92.7
Coding
Claude 5.5 Opus Thinking Max Effort 89.3
Agentic coding
DeepSeek V4.1 Flash Max Effort 77.3
Mathematics
Claude 5.5 Opus Thinking Max Effort 97.1
Data analysis
GPT-6 Astra Max Effort 83.0
Language
Claude Fable 5 Max Effort 90.7
Instruction following
Gemini 3.8 Flash High 81.4

Every number above comes from the 2026-06-25 release dated 2026-06-25, published by a third-party benchmark and reproduced unaltered.

Who is strongest on the leaderboard below.

The frontier chart at the top of the page answers which model is worth paying for. This chart ranks the models in the leaderboard snapshot by score, so you can see how far apart the leaders really are.

How they rank

Longer is better. Bars show the Overall score out of 100; the figure beside each bar is the exact value. Select a row to carry that model into the cost chart below.

Showing the top 12 of 58. Switch to the full comparison view for every model.

Category profile

Pick up to three models from the leaderboard to compare their seven category scores.

  • Claude Fable 5.1 Max Effort83.4
  • ox-alpha-max69.2
  • DeepSeek V4.1 Flash Max Effort81.1
  • Reasoning91.776.686.7
  • Coding86.475.880.0
  • Agentic coding66.152.677.3
  • Mathematics97.077.593.3
  • Data analysis80.375.879.3
  • Language89.566.181.2
  • Instruction following73.060.370.0

Every track runs 0 to 100. Further right is stronger; the bold figure is the leader in that row.

Read these scores as a table
Category profile
CategoryClaude Fable 5.1 Max Effortox-alpha-maxDeepSeek V4.1 Flash Max Effort
Reasoning91.776.686.7
Coding86.475.880.0
Agentic coding66.152.677.3
Mathematics97.077.593.3
Data analysis80.375.879.3
Language89.566.181.2
Instruction following73.060.370.0

Leaderboard

Sort every category, narrow the list, and open a model for its published subtask breakdown.

  1. 01

    Claude Fable 5.1 Max Effort

    Anthropic

    83.4

    Overall

    $1.212

    per successful task

    Mathematics97.0Reasoning91.7Language89.5
  2. 02

    Claude 5.5 Opus Thinking Max Effort

    Anthropic

    83.2

    Overall

    $0.799

    per successful task

    Mathematics97.1Reasoning92.2Coding89.3
  3. 03

    Claude Fable 5 Max Effort

    Anthropic

    83.0

    Overall

    $1.439

    per successful task

    Mathematics96.0Language90.7Reasoning89.7
  4. 04

    GPT-6 Astra Max Effort

    OpenAI

    82.2

    Overall

    $0.736

    per successful task

    Mathematics96.8Reasoning92.7Language89.4
  5. 05

    GPT-6.1 Sol Max Effort

    OpenAI

    81.6

    Overall

    $0.142

    per successful task

    Mathematics96.8Reasoning92.6Language90.1
  6. 06

    Muse Spark 1.3 xHigh Effort

    Meta

    81.6

    Overall

    $0.219

    per successful task

    Mathematics95.9Reasoning89.7Language82.8
  7. 07

    DeepSeek V4.1 Flash Max Effort

    open

    DeepSeek

    81.1

    Overall

    $0.029

    per successful task

    Mathematics93.3Reasoning86.7Language81.2
  8. 08

    GPT-5.6 Sol Max Effort

    OpenAI

    81.0

    Overall

    $0.515

    per successful task

    Mathematics96.2Reasoning91.7Language87.7
  9. 09

    GPT-5.5 Thinking xHigh Effort

    OpenAI

    80.2

    Overall

    $0.435

    per successful task

    Mathematics95.9Reasoning89.7Language87.4
  10. 10

    Claude 5 Opus Thinking Max Effort

    Anthropic

    80.1

    Overall

    $0.699

    per successful task

    Mathematics95.7Reasoning91.2Language88.7
  11. 11

    GPT-6 Sol Max Effort

    OpenAI

    79.3

    Overall

    $0.268

    per successful task

    Mathematics96.4Reasoning88.7Language85.3
  12. 12

    Kimi K3

    open

    Moonshot AI

    79.2

    Overall

    $0.348

    per successful task

    Reasoning90.7Language85.5Mathematics84.4
  13. 13

    Gemini 3.7 Flash High

    Google

    78.8

    Overall

    $0.157

    per successful task

    Mathematics93.5Reasoning87.8Language85.5
  14. 14

    Qwen 3.8 Max

    open

    Alibaba

    78.5

    Overall

    $0.275

    per successful task

    Mathematics91.3Reasoning88.2Language79.7
  15. 15

    Muse Spark 1.2 xHigh Effort

    Meta

    78.0

    Overall

    $0.375

    per successful task

    Mathematics91.2Reasoning90.0Language78.6
  16. 16

    Grok 4.6 xHigh

    xAI

    78.0

    Overall

    $0.207

    per successful task

    Mathematics92.6Reasoning90.5Language83.7
  17. 17

    GPT-5.4 Thinking xHigh Effort

    OpenAI

    78.0

    Overall

    $0.387

    per successful task

    Mathematics94.1Reasoning88.1Language82.6
  18. 18

    GPT-5.6 Terra Max Effort

    OpenAI

    77.9

    Overall

    $0.352

    per successful task

    Mathematics94.9Reasoning90.6Language82.9
  19. 19

    Claude Sonnet 5.5 xHigh Effort

    Anthropic

    77.8

    Overall

    $0.141

    per successful task

    Mathematics96.7Coding88.9Reasoning86.8
  20. 20

    DeepSeek V4 Pro 0813

    open

    DeepSeek

    77.4

    Overall

    $0.044

    per successful task

    Mathematics95.1Reasoning85.8Language82.1
  21. 21

    Grok 4.7 xHigh

    xAI

    77.4

    Overall

    $0.718

    per successful task

    Mathematics95.7Reasoning82.7Language80.1
  22. 22

    Gemini 3.1 Pro Preview High

    Google

    77.0

    Overall

    $0.286

    per successful task

    Mathematics91.0Language85.4Reasoning84.0
  23. 23

    DeepSeek V4 Flash Vision Exp

    open

    DeepSeek

    76.8

    Overall

    $0.051

    per successful task

    Mathematics87.8Reasoning85.4Language80.4
  24. 24

    Claude 4.7 Opus Thinking xHigh Effort

    Anthropic

    76.5

    Overall

    $0.528

    per successful task

    Mathematics92.9Reasoning87.2Coding82.1
  25. 25

    Claude 4.8 Opus Thinking Max Effort

    Anthropic

    76.2

    Overall

    $0.983

    per successful task

    Mathematics94.3Reasoning89.2Coding81.8
  26. 26

    Qwen 3.8 Flash Next

    open

    Alibaba

    76.2

    Overall

    $0.042

    per successful task

    Reasoning87.4Mathematics85.8Instruction following77.1
  27. 27

    GLM-5.3

    open

    Z.AI

    76.1

    Overall

    $0.450

    per successful task

    Mathematics87.9Reasoning85.8Language79.9
  28. 28

    Claude Sonnet 5 xHigh Effort

    Anthropic

    76.0

    Overall

    $0.505

    per successful task

    Mathematics92.9Reasoning88.7Coding80.7
  29. 29

    Grok 4.5

    xAI

    75.8

    Overall

    $0.131

    per successful task

    Mathematics90.8Reasoning87.2Language82.8
  30. 30

    Gemini 3.8 Flash High

    Google

    75.8

    Overall

    $0.307

    per successful task

    Mathematics91.6Reasoning89.3Language87.8
  31. 31

    Qwen3.8 27B

    open

    Alibaba

    75.3

    Overall

    $0.094

    per successful task

    Mathematics86.2Reasoning80.0Data analysis76.6
  32. 32

    Muse Spark 1.1 xHigh Effort

    Meta

    75.3

    Overall

    $0.198

    per successful task

    Reasoning87.7Mathematics87.1Coding77.2
  33. 33

    GPT-5.2 High

    OpenAI

    74.6

    Overall

    $0.234

    per successful task

    Mathematics93.2Reasoning83.2Language79.8
  34. 34

    Gemini 3.5 Flash High

    Google

    74.6

    Overall

    $0.249

    per successful task

    Mathematics88.2Language84.6Reasoning82.0
  35. 35

    Claude 4.6 Opus Thinking High Effort

    Anthropic

    74.5

    Overall

    $0.404

    per successful task

    Mathematics89.3Reasoning88.7Language83.3
  36. 36

    DeepSeek V4 Flash 0731

    open

    DeepSeek

    74.2

    Overall

    $0.060

    per successful task

    Mathematics86.8Reasoning86.6Data analysis79.3
  37. 37

    GPT-5.2 Codex

    OpenAI

    74.0

    Overall

    $0.187

    per successful task

    Mathematics88.8Coding83.6Data analysis78.2
  38. 38

    GPT-5.6 Luna Max Effort

    OpenAI

    73.6

    Overall

    $0.169

    per successful task

    Mathematics87.2Reasoning85.6Coding82.9
  39. 39

    Gemini 3.6 Flash High

    Google

    73.6

    Overall

    $0.235

    per successful task

    Mathematics86.4Reasoning85.1Language83.9
  40. 40

    GLM-5.2

    open

    Z.AI

    73.2

    Overall

    $0.225

    per successful task

    Mathematics89.8Coding79.7Reasoning78.6
  41. 41

    Qwen 3.7 Max

    Alibaba

    73.1

    Overall

    $0.182

    per successful task

    Mathematics85.2Reasoning83.3Language79.7
  42. 42

    Claude 4.6 Sonnet Thinking Medium Effort

    Anthropic

    73.0

    Overall

    $0.306

    per successful task

    Mathematics87.0Reasoning84.8Coding79.3
  43. 43

    Claude 4.5 Opus Thinking High Effort

    Anthropic

    72.6

    Overall

    $0.610

    per successful task

    Mathematics90.4Language81.3Reasoning80.1
  44. 44

    GPT-6 Luna Max Effort

    OpenAI

    72.0

    Overall

    $0.026

    per successful task

    Mathematics89.1Reasoning81.8Coding79.0
  45. 45

    Inkling xHigh Effort

    open

    Thinking Machines

    71.9

    Overall

    $0.310

    per successful task

    Mathematics88.4Reasoning78.3Language73.5
  46. 46

    GLM-5.3 Flash

    open

    Z.AI

    71.6

    Overall

    $0.031

    per successful task

    Mathematics81.2Coding79.0Reasoning77.6
  47. 47

    Kimi K2.6 Thinking

    open

    Moonshot AI

    70.5

    Overall

    $0.169

    per successful task

    Mathematics84.3Reasoning79.4Coding78.6
  48. 48

    GPT-5.4 Nano xHigh

    OpenAI

    69.6

    Overall

    $0.091

    per successful task

    Mathematics91.0Reasoning81.1Coding70.8
  49. 49

    ox-alpha-max

    Stealth

    69.2

    Overall

    $0.000

    per successful task

    Mathematics77.5Reasoning76.6Coding75.8
  50. 50

    Qwen 3.6 Plus

    Alibaba

    68.9

    Overall

    $0.227

    per successful task

    Mathematics83.7Coding78.2Reasoning75.8
  51. 51

    Kimi K2.7 Code

    open

    Moonshot AI

    68.4

    Overall

    $0.100

    per successful task

    Reasoning82.8Mathematics79.6Language77.9
  52. 52

    Grok Build 0.1

    xAI

    67.8

    Overall

    $0.024

    per successful task

    Mathematics78.4Reasoning76.4Language72.5
  53. 53

    Nemotron 3 Ultra 550B A55B

    open

    NVIDIA

    67.4

    Overall

    $0.371

    per successful task

    Mathematics88.7Reasoning74.7Instruction following73.4
  54. 54

    Minimax M3

    Minimax

    67.3

    Overall

    $0.060

    per successful task

    Mathematics76.9Language76.8Data analysis76.2
  55. 55

    GPT-5.4 Mini xHigh

    OpenAI

    66.4

    Overall

    $0.334

    per successful task

    Mathematics78.5Coding71.6Reasoning71.3
  56. 56

    Qwen 3.6 27B

    open

    Alibaba

    64.0

    Overall

    $0.202

    per successful task

    Mathematics79.9Coding71.8Data analysis70.4
  57. 57

    Gemini 3.5 Flash-Lite High

    Google

    63.9

    Overall

    $0.069

    per successful task

    Coding76.1Mathematics73.7Language71.8
  58. 58

    Grok 4.3

    xAI

    62.3

    Overall

    $0.061

    per successful task

    Mathematics84.3Language73.6Reasoning70.8

Cost is the source-provided cost-per-successful-task metric for the selected scope.

Scores are a snapshot of a public third-party benchmark, reproduced unaltered. Dapols adds the cost-per-successful-task analysis on top.View source

A benchmark rank is not a hiring decision.

These scores say which model is strongest at a set of held-out tasks. They don’t say which one fits your workflow, your budget, or the tools you already pay for — and the cheapest model that clears your bar usually beats the highest scorer.

What changed recently

Dated releases from every major lab, with the vendor's own claims labelled as claims.

Open

What it costs to run

Our tool catalogue carries the price we last verified and the date we checked it.

Open

What to actually build

A deployment plan picks the stack for one workflow at your budget — models included.

Open