Acropolis

Decision recommender

Model picker

Filter candidate models by domain focus, cost ceiling, open-weights requirement, and evidence coverage grade, then weight quality, cost efficiency, and reliability to see ranked recommendations with transparent caveats. Data build 17 Nov 2026.

Target domains (leave empty for overall capability)

Applies to models reporting USD task cost.

55
25
20
262 models satisfy your filters
  1. #1

    DeepSeek V4.1 Flash 1017

    DeepSeekOpen weights17 Oct 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score85.4
    Quality: 82 (scale 0 to 100)Quality82
    Cost efficiency: 84 (scale 0 to 100)Cost efficiency84
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 82.36/100 across 15 boards
    • On Pareto cost frontier on 6 boards
    • Open weights
    • Lowest observed run cost: $0.14
  2. #2

    Muse Ember

    Meta9 Nov 2026
    Grade B. Evidence grade B of A to D: good.
    Fit score84.2
    Quality: 82 (scale 0 to 100)Quality82
    Cost efficiency: 100 (scale 0 to 100)Cost efficiency100
    Evidence reliability: 72 (scale 0 to 100)Evidence reliability72

    Why it ranks here

    • Latent capability 81.45/100 across 11 boards
    • On Pareto cost frontier on 10 boards
    • Lowest observed run cost: $0.02
  3. #3

    Gemini 4 Carbon

    Google10 Nov 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score82.1
    Quality: 100 (scale 0 to 100)Quality100
    Cost efficiency: 32 (scale 0 to 100)Cost efficiency32
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 100/100 across 18 boards
    • On Pareto cost frontier on 7 boards
    • Lowest observed run cost: $1.35
  4. #4

    DeepSeek V4.1 Flash

    DeepSeekOpen weights9 Sept 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score80.1
    Quality: 72 (scale 0 to 100)Quality72
    Cost efficiency: 89 (scale 0 to 100)Cost efficiency89
    Evidence reliability: 92 (scale 0 to 100)Evidence reliability92

    Why it ranks here

    • Latent capability 71.62/100 across 9 boards
    • On Pareto cost frontier on 1 boards
    • Open weights
    • Lowest observed run cost: $0.31
  5. #5

    MiMo-V2.6-Flash

    Xiaomi21 Sept 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score80.1
    Quality: 69 (scale 0 to 100)Quality69
    Cost efficiency: 95 (scale 0 to 100)Cost efficiency95
    Evidence reliability: 92 (scale 0 to 100)Evidence reliability92

    Why it ranks here

    • Latent capability 68.81/100 across 16 boards
    • On Pareto cost frontier on 2 boards
    • Lowest observed run cost: $0.05

    Caveats

    • High cross-benchmark variance
  6. #6

    GPT-6.1 Sol

    OpenAI29 Sept 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score79.3
    Quality: 83 (scale 0 to 100)Quality83
    Cost efficiency: 58 (scale 0 to 100)Cost efficiency58
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 83.14/100 across 22 boards
    • On Pareto cost frontier on 5 boards
    • Lowest observed run cost: $0.37
  7. #7

    Muse Wonder Full 1.0

    Meta12 Oct 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score78.7
    Quality: 91 (scale 0 to 100)Quality91
    Cost efficiency: 41 (scale 0 to 100)Cost efficiency41
    Evidence reliability: 92 (scale 0 to 100)Evidence reliability92

    Why it ranks here

    • Latent capability 90.91/100 across 12 boards
    • On Pareto cost frontier on 3 boards
    • Lowest observed run cost: $2.15
  8. #8

    Gemini 4 Hydrogen

    Google12 Nov 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score78.5
    Quality: 91 (scale 0 to 100)Quality91
    Cost efficiency: 40 (scale 0 to 100)Cost efficiency40
    Evidence reliability: 92 (scale 0 to 100)Evidence reliability92

    Why it ranks here

    • Latent capability 91.1/100 across 15 boards
    • On Pareto cost frontier on 6 boards
    • Lowest observed run cost: $0.28
  9. #9

    MiMo-V2.6-Pro

    Xiaomi21 Sept 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score77.6
    Quality: 74 (scale 0 to 100)Quality74
    Cost efficiency: 73 (scale 0 to 100)Cost efficiency73
    Evidence reliability: 92 (scale 0 to 100)Evidence reliability92

    Why it ranks here

    • Latent capability 74.41/100 across 16 boards
    • Lowest observed run cost: $0.13
  10. #10

    Gemini 4 Argon

    Google13 Oct 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score77.5
    Quality: 93 (scale 0 to 100)Quality93
    Cost efficiency: 29 (scale 0 to 100)Cost efficiency29
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 92.77/100 across 19 boards
    • On Pareto cost frontier on 6 boards
    • Lowest observed run cost: $0.70
  11. #11

    Inkling 2 Small

    Thinking Machines5 Nov 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score77.5
    Quality: 75 (scale 0 to 100)Quality75
    Cost efficiency: 72 (scale 0 to 100)Cost efficiency72
    Evidence reliability: 92 (scale 0 to 100)Evidence reliability92

    Why it ranks here

    • Latent capability 74.84/100 across 14 boards
    • On Pareto cost frontier on 2 boards
    • Lowest observed run cost: $0.14

    Caveats

    • High cross-benchmark variance
  12. #12

    Fable 5.2

    Anthropic9 Nov 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score77.1
    Quality: 95 (scale 0 to 100)Quality95
    Cost efficiency: 22 (scale 0 to 100)Cost efficiency22
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 95.25/100 across 15 boards
    • On Pareto cost frontier on 7 boards
    • Lowest observed run cost: $3.31
  13. #13

    GPT-6.1 Terra

    OpenAI26 Oct 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score76.6
    Quality: 76 (scale 0 to 100)Quality76
    Cost efficiency: 62 (scale 0 to 100)Cost efficiency62
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 76.11/100 across 14 boards
    • On Pareto cost frontier on 4 boards
    • Lowest observed run cost: $0.16

    Caveats

    • High cross-benchmark variance
  14. #14

    GPT-6 Luna

    OpenAI22 Sept 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score76.1
    Quality: 62 (scale 0 to 100)Quality62
    Cost efficiency: 91 (scale 0 to 100)Cost efficiency91
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 62.36/100 across 21 boards
    • On Pareto cost frontier on 2 boards
    • Lowest observed run cost: $0.01

    Caveats

    • High cross-benchmark variance
  15. #15

    Sonnet 5.5

    Anthropic28 Sept 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score75.7
    Quality: 92 (scale 0 to 100)Quality92
    Cost efficiency: 24 (scale 0 to 100)Cost efficiency24
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 91.59/100 across 22 boards
    • On Pareto cost frontier on 5 boards
    • Lowest observed run cost: $0.37
  16. #16

    GPT-5.6 Luna

    OpenAI9 Jul 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score75.7
    Quality: 67 (scale 0 to 100)Quality67
    Cost efficiency: 83 (scale 0 to 100)Cost efficiency83
    Evidence reliability: 92 (scale 0 to 100)Evidence reliability92

    Why it ranks here

    • Latent capability 66.69/100 across 13 boards
    • On Pareto cost frontier on 2 boards
    • Lowest observed run cost: $0.03
  17. #17

    Opus 5.5

    Anthropic22 Sept 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score74.4
    Quality: 92 (scale 0 to 100)Quality92
    Cost efficiency: 17 (scale 0 to 100)Cost efficiency17
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 92.4/100 across 23 boards
    • On Pareto cost frontier on 3 boards
    • Lowest observed run cost: $1.17
  18. #18

    Haiku 5.5

    Anthropic7 Oct 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score73.7
    Quality: 72 (scale 0 to 100)Quality72
    Cost efficiency: 60 (scale 0 to 100)Cost efficiency60
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 71.93/100 across 20 boards
    • On Pareto cost frontier on 2 boards
    • Lowest observed run cost: $0.08
  19. #19

    GLM-5.3 Flash

    Z.aiOpen weights20 Aug 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score73.7
    Quality: 67 (scale 0 to 100)Quality67
    Cost efficiency: 71 (scale 0 to 100)Cost efficiency71
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 66.87/100 across 20 boards
    • Open weights
    • Lowest observed run cost: $0.07

    Caveats

    • High cross-benchmark variance
  20. #20

    Kimi K3.1 Pro

    Moonshot AIOpen weights23 Oct 2026
    Grade A. Evidence grade A of A to D: strongest.
    Fit score72.6
    Quality: 79 (scale 0 to 100)Quality79
    Cost efficiency: 41 (scale 0 to 100)Cost efficiency41
    Evidence reliability: 96 (scale 0 to 100)Evidence reliability96

    Why it ranks here

    • Latent capability 78.51/100 across 15 boards
    • On Pareto cost frontier on 1 boards
    • Open weights
    • Lowest observed run cost: $1.16

    Caveats

    • High cross-benchmark variance