guava Book a call

Main Street Bench

Main Street Bench v0.14 — deepseek-v4-1-flash high DeepSeek V4.1 Flash

Score: 6 / 100

Protected $53,050 of the $100,000 at risk. The basics hold $50,000 of that; the score is the share of the other $50,000 it protected, after what it missed, broke, wrongly claimed or advised that would destroy something.

Fixed 37 of 100 problems: 21 of the 35 basics and 16 of the other 65 (16 checked by code, 0 by the judge). Broke 0 things that worked.

Scorecard // run 1Score · $ for the run

Run 1DeepSeek V4.1 Flash High DeepSeek6$0.25
AreaFixed
Menu, prices and allergens1/4
Website and ordering2/16
Money and books6/18
Policies and legal3/11
Customers and inbox4/16

By how hard they were to spot: Obvious 0/2, Digging 8/27, Hidden 8/36

The basics21 of 35 fixed; the misses cost $11,950
The other 6516 fixed, $15,000 of $50,000 caught
What it brokenothing
Said it fixed something, but didn't0
Advised a step that would destroy something0
Judgegrok-4.7 (high) via Cursor
Model the tool reportedDeepSeek V4.1 Flash High, deepseek-v4-1-flash-high
Tool profilethe person's own settings
Integrityclean (this tool's log doesn't list commands, so only copying was checked); personal instructions loaded: ~/.codex/AGENTS.md
Run time3 min 22 s
Tokens burned5.93M (55.7K output)
Speed276 output tokens per second
Cost$0.25, estimated from token counts at public API prices

Run 2026-10-13-deepseek-v4-1-flash-high-v0.14-01 from commit 287cd3e on 2026-10-13.

38 runsEach run’s scorecard

  1. Claude Opus 5.581 out of 100Run 1 10 min · $2.50Scorecard
  2. Claude Opus 5.574 out of 100Run 2 8 min · $2.31Scorecard
  3. Claude Opus 5.571 out of 100Run 3 12 min · $2.78Scorecard
  4. GPT-6.1 Sol69 out of 100Run 1 16 min · ~$0.70Scorecard
  5. GPT-6.1 Sol75 out of 100Run 2 15 min · ~$0.77Scorecard
  6. GPT-6.1 Sol72 out of 100Run 3 12 min · ~$0.63Scorecard
  7. GPT-6 Astra55 out of 100Run 1 12 min · ~$3.52Scorecard
  8. GPT-6 Astra80 out of 100Run 2 13 min · ~$3.99Scorecard
  9. GPT-6 Astra79 out of 100Run 3 12 min · ~$3.38Scorecard
  10. Gemini 4 Argon56 out of 100Run 1 14 min · ~$4.17Scorecard
  11. Gemini 4 Argon50 out of 100Run 2 10 min · ~$2.36Scorecard
  12. Gemini 4 Argon58 out of 100Run 3 16 min · ~$4.02Scorecard
  13. Claude Sonnet 5.564 out of 100Run 1 8 min · $1.52Scorecard
  14. Claude Sonnet 5.522 out of 100Run 2 8 min · $1.53Scorecard
  15. Claude Sonnet 5.562 out of 100Run 3 8 min · $1.72Scorecard
  16. Grok 4.746 out of 100Run 1 18 min · ~$3.32Scorecard
  17. Grok 4.746 out of 100Run 2 19 min · —Scorecard
  18. Grok 4.753 out of 100Run 3 20 min · ~$3.28Scorecard
  19. Muse Wonder49 out of 100Run 1 7 min · ~$3.45Scorecard
  20. Muse Wonder45 out of 100Run 2 6 min · ~$2.76Scorecard
  21. Muse Wonder35 out of 100Run 3 30 min · ~$2.34Scorecard
  22. Muse Wonder0 out of 100Run 4 38 min · ~$2.07Scorecard
  23. GLM 5.329 out of 100Run 1 8 min · ~$2.15Scorecard
  24. GLM 5.328 out of 100Run 2 18 min · ~$1.29Scorecard
  25. GLM 5.317 out of 100Run 3 19 min · ~$1.95Scorecard
  26. Gemini 3.8 Flash14 out of 100Run 1 11 min · ~$1.19Scorecard
  27. Gemini 3.8 Flash15 out of 100Run 2 9 min · ~$1.06Scorecard
  28. Gemini 3.8 Flash21 out of 100Run 3 10 min · ~$1.23Scorecard
  29. Kimi K30 out of 100Run 1 4 min · ~$2.60Scorecard
  30. Kimi K334 out of 100Run 2 21 min · ~$1.46Scorecard
  31. Kimi K331 out of 100Run 3 21 min · ~$1.45Scorecard
  32. DeepSeek V4.1 Flash6 out of 100Run 1 3 min · ~$0.25This scorecard
  33. DeepSeek V4.1 Flash9 out of 100Run 2 4 min · ~$0.13Scorecard
  34. DeepSeek V4.1 Flash16 out of 100Run 3 8 min · ~$0.28Scorecard
  35. Muse Spark Open17 out of 100Run 1 17 min · $2.47Scorecard
  36. Muse Spark Open0 out of 100Run 2 30 min · $2.78Scorecard
  37. Muse Spark Open0 out of 100Run 3 30 min · $8.25Scorecard
  38. Muse Spark Open0 out of 100Run 4 28 min · $3.54Scorecard