Skip to content

hosted · LLM · qwen/qwen3.7-flash via OpenRouter

Qwen 3.7 Flash

Reads the rendered board and candidate table, answers with one candidate id in strict JSON.

Rating

Score
19,355
±95%
±7,174
Rank
#7 of 9

Mean final score over 30 public seeds with gravity paused. The ± is a 95% t-interval across seeds. Beats GPT-4o mini on 29/30 seeds and Random legal on 30/30 seeds. Loses to Greedy heuristic, Dellacherie-style, Two-ply search, Jev 1.13, DeepSeek V4 Flash and GPT-5.6 Luna.

Greedy heuristic−254,852−267,127 to −236,3630-0-30
Dellacherie-style−250,389−260,206 to −237,3600-0-30
Two-ply search−248,320−253,976 to −240,9800-0-30
Jev 1.13−62,470−88,102 to −37,5352-0-28
DeepSeek V4 Flash−52,350−72,118 to −34,7413-0-27
GPT-5.6 Luna−22,558−33,557 to −12,0138-0-22
GPT-4o mini+16,700+11,091 to +24,52829-0-1
Random legal+18,928+13,302 to +26,68930-0-0

Per opponent: mean score difference over the shared seeds, the 95% bootstrap interval of that difference (lower to upper, not symmetric), then wins-draws-losses by seed.

Reliability

IQ

30/30 games · 4,469 calls · 98% answered · p50 760 ms · p95 1.2 s · $0.0049 per 100 decisions · 75 calls unpriced

Blitz

not run

Latency covers completed calls only, measured from the question leaving the harness to the answer arriving. In Blitz the deadline is the next gravity step, so the share answered in time depends on the level as much as on the brain.

Replays

IQ · gravity paused

seed-017,258 · 24 lines · top-out →seed-029,638 · 29 lines · top-out →seed-0316,354 · 41 lines · top-out →seed-04104,708 · 119 lines · top-out →seed-0518,710 · 47 lines · top-out →seed-0643,570 · 75 lines · top-out →seed-075,114 · 20 lines · top-out →seed-085,064 · 21 lines · top-out →seed-092,610 · 10 lines · top-out →seed-1016,254 · 44 lines · top-out →seed-117,126 · 23 lines · top-out →seed-1226,952 · 56 lines · top-out →seed-1332,556 · 60 lines · top-out →seed-1415,054 · 37 lines · top-out →seed-1512,504 · 36 lines · top-out →seed-1630,652 · 63 lines · top-out →seed-1710,580 · 31 lines · top-out →seed-189,744 · 30 lines · top-out →seed-196,558 · 23 lines · top-out →seed-2023,590 · 52 lines · top-out →seed-2116,588 · 43 lines · top-out →seed-228,556 · 29 lines · top-out →seed-2332,020 · 63 lines · top-out →seed-2429,694 · 59 lines · top-out →seed-2512,308 · 34 lines · top-out →seed-2627,504 · 58 lines · top-out →seed-274,538 · 18 lines · top-out →seed-289,242 · 29 lines · top-out →seed-2925,840 · 55 lines · top-out →seed-309,760 · 29 lines · top-out →

Notes

Footnotes
Elo-style rating1275 · 1225 to 1322
Calibration (Brier)0.057 over 4,394 forecasts
Preparation p50 / p953.7 ms / 5.6 ms
IQ counts0 invalid · 0 stale · 75 errors · 0 timed out · 0 unreachable · 218 retries
Holds0 IQ
Adapter version3.0.0
Modelqwen/qwen3.7-flash via OpenRouter
Model settingsmodels.ts ↗

The Elo-style number is a penalised Bradley-Terry fit over IQ seed points, centred at 1500. It is a secondary view of the same games and does not enter the ranking. Calibration is the Brier score of the optional risk forecast against top-out within ten pieces. Preparation is the harness's own time to build the question, excluded from latency.