← arena · results

analysis

4 models · 7,561 rolls · $5.55

the average race

Average victory points at each fraction of game progress, all 50 games. The gold band spans the strongest to weakest seat at every step.

0510avg VPsstarthalfgame overgemini-3.7-flashgemini-3.7-flash 5.7deepseek-v4-flashdeepseek-v4-flash 6.94gpt-5.6-lunagpt-5.6-luna 7.42hy3hy3 5.04

won from behind

Winners who trailed badly mid-game, worst dip first.

0.5129% of leader pace at 37% through the gamestartgame overshare of leader
deepseek-v4-flash m020_g0
fell to 29% of leader pace - won with 10 VPs
0.5150% of leader pace at 2% through the gamestartgame overshare of leader
deepseek-v4-flash m006_g0
fell to 50% of leader pace - won with 10 VPs
0.5150% of leader pace at 37% through the gamestartgame overshare of leader
deepseek-v4-flash m029_g0
fell to 50% of leader pace - won with 10 VPs

how close games were

Gap between winner and runner-up across all 50 games - most titles were decided late.

481216margin of 1 VPs: 7 games71margin of 2 VPs: 16 games162margin of 3 VPs: 9 games93margin of 4 VPs: 5 games54margin of 5 VPs: 6 games65margin of 6 VPs: 6 games66margin of 7 VPs: 1 games17median margin 3 VPsmedian 3winner's margin in VPs

game length by winner

How long each model's wins took - gold tick marks the row median.

100200300400turnsdeepseek-v4-flashdeepseek-v4-flash won in 102 turnsdeepseek-v4-flash won in 105 turnsdeepseek-v4-flash won in 109 turnsdeepseek-v4-flash won in 110 turnsdeepseek-v4-flash won in 139 turnsdeepseek-v4-flash won in 146 turnsdeepseek-v4-flash won in 151 turnsdeepseek-v4-flash won in 160 turnsdeepseek-v4-flash won in 180 turnsdeepseek-v4-flash won in 183 turnsdeepseek-v4-flash won in 184 turnsdeepseek-v4-flash won in 187 turnsdeepseek-v4-flash won in 198 turnsdeepseek-v4-flash won in 207 turnsdeepseek-v4-flash won in 214 turnsdeepseek-v4-flash won in 218 turnsdeepseek-v4-flash won in 241 turnsdeepseek-v4-flash won in 248 turnsdeepseek-v4-flash won in 355 turnsdeepseek-v4-flash median 183 turnsmed 183gpt-5.6-lunagpt-5.6-luna won in 70 turnsgpt-5.6-luna won in 73 turnsgpt-5.6-luna won in 91 turnsgpt-5.6-luna won in 100 turnsgpt-5.6-luna won in 100 turnsgpt-5.6-luna won in 106 turnsgpt-5.6-luna won in 113 turnsgpt-5.6-luna won in 127 turnsgpt-5.6-luna won in 128 turnsgpt-5.6-luna won in 129 turnsgpt-5.6-luna won in 132 turnsgpt-5.6-luna won in 137 turnsgpt-5.6-luna won in 147 turnsgpt-5.6-luna won in 150 turnsgpt-5.6-luna won in 154 turnsgpt-5.6-luna won in 157 turnsgpt-5.6-luna won in 165 turnsgpt-5.6-luna won in 204 turnsgpt-5.6-luna won in 214 turnsgpt-5.6-luna median 129 turnsmed 129gemini-3.7-flashgemini-3.7-flash won in 58 turnsgemini-3.7-flash won in 131 turnsgemini-3.7-flash won in 156 turnsgemini-3.7-flash won in 194 turnsgemini-3.7-flash won in 194 turnsgemini-3.7-flash won in 212 turnsgemini-3.7-flash won in 234 turnsgemini-3.7-flash median 194 turnsmed 194hy3hy3 won in 114 turnshy3 won in 122 turnshy3 won in 134 turnshy3 won in 143 turnshy3 won in 189 turnshy3 median 134 turnsmed 134

robber diplomacy — the leader gets robbed

deepseekgpt-lunageminihy3deepseekgpt-lunageminihy3a model never robs itself–deepseek-v4-flash robbed gpt-5.6-luna 190 times190deepseek-v4-flash robbed gemini-3.7-flash 152 times152deepseek-v4-flash robbed hy3 112 times112gpt-5.6-luna robbed deepseek-v4-flash 240 times240a model never robs itself–gpt-5.6-luna robbed gemini-3.7-flash 159 times159gpt-5.6-luna robbed hy3 117 times117gemini-3.7-flash robbed deepseek-v4-flash 180 times180gemini-3.7-flash robbed gpt-5.6-luna 174 times174a model never robs itself–gemini-3.7-flash robbed hy3 83 times83hy3 robbed deepseek-v4-flash 168 times168hy3 robbed gpt-5.6-luna 137 times137hy3 robbed gemini-3.7-flash 101 times101a model never robs itself–
everyone's favourite target is a co-leader — 60% of all robberies land on deepseek or gpt-luna. Gold outline marks each robber's top victim.

play styles — four different brains

ROBBERDEVKNIGHTSTRADESACTIVEdeepseek-v4-flash19%5.322.6413.195%2416 movesdeepseek-v4-flashROBBERDEVKNIGHTSTRADESACTIVEgpt-5.6-luna20%7.043.9212.9897%2535 movesgpt-5.6-lunaROBBERDEVKNIGHTSTRADESACTIVEgemini-3.7-flash18%4.562.721084%2366 movesgemini-3.7-flashROBBERDEVKNIGHTSTRADESACTIVEhy319%3.361.7410.7881%2167 moveshy3
spokes normalized to the best model per axis, raw values on the vertices. luna buys dev cards and knights, deepseek deals the most, hy3 sits on its hands — passing a fifth of its turns.

what they trade — wheat out, ore in

GIVEGETdeepseek-v4-flash gave wheat away 161 timesdeepseek-v4-flash took wheat 137 timesdeepseek-v4-flash gave sheep away 114 timesdeepseek-v4-flash took sheep 28 timesdeepseek-v4-flash gave wood away 126 timesdeepseek-v4-flash took wood 51 timesdeepseek-v4-flash gave ore away 82 timesdeepseek-v4-flash took ore 188 timesdeepseek-v4-flash gave brick away 54 timesdeepseek-v4-flash took brick 133 timesgpt-5.6-luna gave wheat away 115 timesgpt-5.6-luna took wheat 153 timesgpt-5.6-luna gave sheep away 146 timesgpt-5.6-luna took sheep 16 timesgpt-5.6-luna gave wood away 77 timesgpt-5.6-luna took wood 31 timesgpt-5.6-luna gave ore away 54 timesgpt-5.6-luna took ore 217 timesgpt-5.6-luna gave brick away 70 timesgpt-5.6-luna took brick 45 timesgemini-3.7-flash gave wheat away 89 timesgemini-3.7-flash took wheat 83 timesgemini-3.7-flash gave sheep away 43 timesgemini-3.7-flash took sheep 34 timesgemini-3.7-flash gave wood away 80 timesgemini-3.7-flash took wood 49 timesgemini-3.7-flash gave ore away 73 timesgemini-3.7-flash took ore 83 timesgemini-3.7-flash gave brick away 44 timesgemini-3.7-flash took brick 80 timeshy3 gave wheat away 47 timeshy3 took wheat 104 timeshy3 gave sheep away 71 timeshy3 took sheep 6 timeshy3 gave wood away 62 timeshy3 took wood 24 timeshy3 gave ore away 50 timeshy3 took ore 125 timeshy3 gave brick away 63 timeshy3 took brick 34 timesall models gave wheat away 412 timeswheat412all models took wheat 477 timeswheat477all models gave sheep away 374 timessheep374all models took sheep 84 timessheep84all models gave wood away 345 timeswood345all models took wood 155 timeswood155all models gave ore away 259 timesore259all models took ore 613 timesore613all models gave brick away 231 timesbrick231all models took brick 292 timesbrick292deepseek-v4-flash: 537 tradesdeepseekgpt-5.6-luna: 462 tradesgpt-lunagemini-3.7-flash: 329 tradesgeminihy3: 293 tradeshy3
wheat leaves (412 deals) and ore arrives (613) — everyone runs the same engine: sell wheat and sheep, hoard ore for dev cards.

where they settled — the same golden corners

chosen 31 timeschosen 6 timeschosen 9 timeschosen 2 timeschosen 8 timeschosen 2 timeschosen 24 timeschosen 2 timeschosen 25 timeschosen 7 timeschosen 16 timeschosen 6 timeschosen 8 timeschosen 9 timeschosen 20 timeschosen 4 timeschosen 19 timeschosen 5 timeschosen 6 timeschosen 8 timeschosen 18 timeschosen 3 timeschosen 16 timeschosen 32 timeschosen 2 timeschosen 4 timeschosen 15 timeschosen 7 timeschosen 1 timeschosen 11 timeschosen 8 timeschosen 4 timeschosen 22 timeschosen 1 timeschosen 10 timeschosen 2 timeschosen 1 timeschosen 7 timeschosen 1 timeschosen 3 timeschosen 7 timeschosen 3 timeschosen 1 timeschosen 1 timeschosen 1 timeschosen 2 times
every opening settlement from all 50 games overlaid on the canonical board - the models never saw it

value for money

more wins / less $ per win0.000.100.200.300102030405025% = coin flipmean $0.16/winwin rate %deepseek-v4-flash: 38% wins at $0.067/win - $1.27 total spenddeepseek-v4-flashgpt-5.6-luna: 38% wins at $0.064/win - $1.23 total spendgpt-5.6-lunagemini-3.7-flash: 14% wins at $0.204/win - $1.43 total spendgemini-3.7-flashhy3: 10% wins at $0.324/win - $1.62 total spendhy3

Same four opponents, same 50 games — yet a win costs $0.064 for gpt-5.6-luna and $0.324 for hy3: a 5.1x spread. The two winners sit inside the gold corner; everyone else paid more to lose more.

openings lined up with the standings

0.050.100.150.20deepseek-v4-flash: mean 0.2053, median 0.194 cards/turn over 100 settlementsdeepseek-v4-flash0.2053deepseek · 19 winsgpt-5.6-luna: mean 0.1936, median 0.194 cards/turn over 100 settlementsgpt-5.6-luna0.1936gpt · 19 winsgemini-3.7-flash: mean 0.1586, median 0.139 cards/turn over 100 settlementsgemini-3.7-flash0.1586gemini · 7 winshy3: mean 0.1586, median 0.139 cards/turn over 100 settlementshy30.1586hy3 · 5 wins

Opening quality is pip count: each settlement's tiles weighted by their dice-roll probability, averaged over both initial settlements. The order happens to match the standings — but that is four data points, and it does not hold under a different slice: counting only each model's first settlement, gemini (72% on the high-yield nodes) outranks luna (70%) while finishing twelve wins behind. Read this as a description of how they drafted, not as a cause of who won.

fair dice

0%4%8%12%16%2: observed 2.50% vs expected 2.78%2: expected 2.78%23: observed 5.54% vs expected 5.56%3: expected 5.56%34: observed 8.95% vs expected 8.33%4: expected 8.33%45: observed 11.39% vs expected 11.11%5: expected 11.11%56: observed 13.78% vs expected 13.89%6: expected 13.89%68: observed 13.46% vs expected 13.89%8: expected 13.89%89: observed 10.81% vs expected 11.11%9: expected 11.11%910: observed 8.13% vs expected 8.33%10: expected 8.33%1011: observed 5.89% vs expected 5.56%11: expected 5.56%1112: observed 2.72% vs expected 2.78%12: expected 2.78%12observedexpectedtwo-dice sum (7 excluded: it moves the robber)

chi² 9.7 on df 9 (95% crit 16.9) over 7,561 rolls — fair dice. No model can blame the board.

who kept stalling

deepseek-v4-flash minute 0: 52 retriesdeepseek-v4-flash minute 1: 29 retriesdeepseek-v4-flash minute 2: 39 retriesdeepseek-v4-flash minute 3: 54 retriesdeepseek-v4-flash minute 4: 51 retriesdeepseek-v4-flash minute 5: 40 retriesdeepseek-v4-flash minute 6: 46 retriesdeepseek-v4-flash minute 7: 41 retriesdeepseek-v4-flash minute 8: 39 retriesdeepseek-v4-flash minute 9: 39 retriesdeepseek-v4-flash minute 10: 44 retriesdeepseek-v4-flash minute 11: 40 retriesdeepseek-v4-flash minute 12: 47 retriesdeepseek-v4-flash minute 13: 41 retriesdeepseek-v4-flash minute 14: 38 retriesdeepseek-v4-flash minute 15: 40 retriesdeepseek-v4-flash minute 16: 42 retriesdeepseek-v4-flash minute 17: 44 retriesdeepseek-v4-flash minute 18: 42 retriesdeepseek-v4-flash minute 19: 38 retriesdeepseek-v4-flash minute 20: 40 retriesdeepseek-v4-flash minute 21: 40 retriesdeepseek-v4-flash minute 22: 36 retriesdeepseek-v4-flash minute 23: 41 retriesdeepseek-v4-flash minute 24: 41 retriesdeepseek-v4-flash minute 25: 40 retriesdeepseek-v4-flash minute 26: 25 retriesdeepseek-v4-flash minute 27: 44 retriesdeepseek-v4-flash minute 28: 38 retriesdeepseek-v4-flash minute 29: 44 retriesdeepseek-v4-flash minute 30: 33 retriesdeepseek-v4-flash minute 31: 30 retriesdeepseek-v4-flash minute 32: 43 retriesdeepseek-v4-flash minute 33: 41 retriesdeepseek-v4-flash minute 34: 36 retriesdeepseek-v4-flash minute 35: 44 retriesdeepseek-v4-flash minute 36: 34 retriesdeepseek-v4-flash minute 37: 31 retriesdeepseek-v4-flash minute 38: 40 retriesdeepseek-v4-flash minute 39: 38 retriesdeepseek-v4-flash minute 40: 42 retriesdeepseek-v4-flash minute 41: 38 retriesdeepseek-v4-flash minute 42: 35 retriesdeepseek-v4-flash minute 43: 34 retriesdeepseek-v4-flash minute 44: 41 retriesdeepseek-v4-flash minute 45: 37 retriesdeepseek-v4-flash minute 46: 38 retriesdeepseek-v4-flash minute 47: 33 retriesdeepseek-v4-flash minute 48: 33 retriesdeepseek-v4-flash minute 49: 38 retriesdeepseek-v4-flash minute 50: 39 retriesdeepseek-v4-flash minute 51: 42 retriesdeepseek-v4-flash minute 52: 35 retriesdeepseek-v4-flash minute 53: 28 retriesdeepseek-v4-flash minute 54: 32 retriesdeepseek-v4-flash minute 55: 31 retriesdeepseek-v4-flash minute 56: 30 retriesdeepseek-v4-flash minute 57: 31 retriesdeepseek-v4-flash minute 58: 30 retriesdeepseek-v4-flash minute 59: 33 retriesdeepseek-v4-flash minute 60: 32 retriesdeepseek-v4-flash minute 61: 32 retriesdeepseek-v4-flash minute 62: 34 retriesdeepseek-v4-flash minute 63: 28 retriesdeepseek-v4-flash minute 64: 28 retriesdeepseek-v4-flash minute 65: 32 retriesdeepseek-v4-flash minute 66: 36 retriesdeepseek-v4-flash minute 67: 29 retriesdeepseek-v4-flash minute 68: 24 retriesdeepseek-v4-flash minute 69: 39 retriesdeepseek-v4-flash minute 70: 21 retriesdeepseek-v4-flash minute 71: 24 retriesdeepseek-v4-flash minute 72: 22 retriesdeepseek-v4-flash minute 73: 25 retriesdeepseek-v4-flash minute 74: 21 retriesdeepseek-v4-flash minute 75: 30 retriesdeepseek-v4-flash minute 76: 22 retriesdeepseek-v4-flash minute 77: 23 retriesdeepseek-v4-flash minute 78: 27 retriesdeepseek-v4-flash minute 79: 26 retriesdeepseek-v4-flash minute 80: 24 retriesdeepseek-v4-flash minute 81: 24 retriesdeepseek-v4-flash minute 82: 20 retriesdeepseek-v4-flash minute 83: 21 retriesdeepseek-v4-flash minute 84: 28 retriesdeepseek-v4-flash minute 85: 23 retriesdeepseek-v4-flash minute 86: 25 retriesdeepseek-v4-flash minute 87: 21 retriesdeepseek-v4-flash minute 88: 17 retriesdeepseek-v4-flash minute 89: 22 retriesdeepseek-v4-flash minute 90: 20 retriesdeepseek-v4-flash minute 91: 14 retriesdeepseek-v4-flash minute 92: 16 retriesdeepseek-v4-flash minute 93: 8 retriesdeepseek-v4-flash minute 94: 11 retriesdeepseek-v4-flash minute 95: 10 retriesdeepseek-v4-flash minute 96: 20 retriesdeepseek-v4-flash minute 97: 15 retriesdeepseek-v4-flash minute 98: 13 retriesdeepseek-v4-flash minute 99: 19 retriesdeepseek-v4-flash minute 100: 15 retriesdeepseek-v4-flash minute 101: 12 retriesdeepseek-v4-flash minute 102: 8 retriesdeepseek-v4-flash minute 103: 11 retriesdeepseek-v4-flash minute 104: 10 retriesdeepseek-v4-flash minute 105: 14 retriesdeepseek-v4-flash minute 106: 9 retriesdeepseek-v4-flash minute 107: 8 retriesdeepseek-v4-flash minute 108: 9 retriesdeepseek-v4-flash minute 109: 10 retriesdeepseek-v4-flash minute 110: 8 retriesdeepseek-v4-flash minute 111: 7 retriesdeepseek-v4-flash minute 112: 9 retriesdeepseek-v4-flash minute 113: 6 retriesdeepseek-v4-flash minute 114: 10 retriesdeepseek-v4-flash minute 115: 6 retriesdeepseek-v4-flash minute 116: 9 retriesdeepseek-v4-flash minute 117: 8 retriesdeepseek-v4-flash minute 118: 8 retriesdeepseek-v4-flash minute 119: 7 retriesdeepseek-v4-flash minute 120: 5 retriesdeepseek-v4-flash minute 121: 4 retriesdeepseek-v4-flash minute 122: 7 retriesdeepseek-v4-flash minute 123: 8 retriesdeepseek-v4-flash minute 124: 7 retriesdeepseek-v4-flash minute 125: 5 retriesdeepseek-v4-flash minute 126: 8 retriesdeepseek-v4-flash minute 127: 6 retriesdeepseek-v4-flash minute 128: 4 retriesdeepseek-v4-flash minute 129: 5 retriesdeepseek-v4-flash minute 130: 5 retriesdeepseek-v4-flash minute 131: 3 retriesdeepseek-v4-flash minute 132: 5 retriesdeepseek-v4-flash minute 133: 3 retriesdeepseek-v4-flash minute 134: 4 retriesdeepseek-v4-flash minute 135: 4 retriesdeepseek-v4-flash minute 136: 3 retriesdeepseek-v4-flash minute 137: 6 retriesdeepseek-v4-flash minute 138: 1 retriesdeepseek-v4-flash minute 139: 7 retriesdeepseek-v4-flash minute 140: 5 retriesdeepseek-v4-flash minute 141: 3 retriesdeepseek-v4-flash minute 142: 4 retriesdeepseek-v4-flash minute 143: 6 retriesdeepseek-v4-flash minute 144: 3 retriesdeepseek-v4-flash minute 145: 3 retriesdeepseek-v4-flash minute 146: 6 retriesdeepseek-v4-flash minute 147: 6 retriesdeepseek-v4-flash minute 148: 5 retriesdeepseek-v4-flash minute 149: 4 retriesdeepseek-v4-flash minute 150: 5 retriesdeepseek-v4-flash minute 151: 4 retriesdeepseek-v4-flash minute 152: 4 retriesdeepseek-v4-flash minute 153: 3 retriesdeepseek-v4-flash minute 154: 5 retriesdeepseek-v4-flash minute 155: 4 retriesdeepseek-v4-flash minute 156: 4 retriesdeepseek-v4-flash minute 157: 1 retriesdeepseek-v4-flash minute 158: 3 retriesdeepseek-v4-flash minute 159: 2 retriesdeepseek-v4-flash minute 160: 2 retriesdeepseek-v4-flash minute 161: 3 retriesdeepseek-v4-flash minute 162: 4 retriesdeepseek-v4-flash minute 163: 2 retriesdeepseek-v4-flash minute 164: 1 retriesdeepseek-v4-flash minute 166: 2 retriesdeepseek-v4-flash minute 167: 1 retriesdeepseek-v4-flash minute 168: 1 retriesdeepseek-v4-flash minute 169: 2 retriesdeepseek-v4-flash minute 170: 3 retriesdeepseek-v4-flash minute 171: 2 retriesdeepseek-v4-flash minute 172: 2 retriesdeepseek-v4-flash minute 173: 2 retriesdeepseek-v4-flash minute 174: 1 retriesdeepseek-v4-flash minute 176: 1 retriesdeepseek-v4-flash minute 177: 1 retriesdeepseek-v4-flash minute 178: 1 retriesdeepseek-v4-flash minute 179: 1 retriesdeepseek-v4-flash minute 181: 1 retriesdeepseek-v4-flash minute 183: 1 retriesdeepseek-v4-flash minute 185: 1 retriesdeepseek-v4-flash minute 187: 1 retriesdeepseek-v4-flash minute 190: 2 retriesdeepseek-v4-flash minute 191: 1 retriesdeepseek-v4-flash minute 193: 1 retriesdeepseek-v4-flash3,635retriesgpt-5.6-luna minute 24: 1 retriesgpt-5.6-luna minute 71: 1 retriesgpt-5.6-luna2gemini-3.7-flash minute 71: 3 retriesgemini-3.7-flash3hy3 minute 0: 12 retrieshy3 minute 1: 20 retrieshy3 minute 2: 22 retrieshy3 minute 3: 9 retrieshy3 minute 4: 4 retrieshy3 minute 5: 6 retrieshy3 minute 6: 5 retrieshy3 minute 7: 4 retrieshy3 minute 8: 8 retrieshy3 minute 9: 4 retrieshy3 minute 10: 8 retrieshy3 minute 11: 4 retrieshy3 minute 12: 3 retrieshy3 minute 13: 5 retrieshy3 minute 14: 4 retrieshy3 minute 15: 5 retrieshy3 minute 16: 8 retrieshy3 minute 17: 5 retrieshy3 minute 18: 9 retrieshy3 minute 19: 12 retrieshy3 minute 20: 13 retrieshy3 minute 21: 9 retrieshy3 minute 22: 5 retrieshy3 minute 23: 5 retrieshy3 minute 24: 4 retrieshy3 minute 25: 14 retrieshy3 minute 26: 15 retrieshy3 minute 27: 9 retrieshy3 minute 28: 6 retrieshy3 minute 29: 7 retrieshy3 minute 30: 16 retrieshy3 minute 31: 16 retrieshy3 minute 32: 6 retrieshy3 minute 33: 6 retrieshy3 minute 34: 5 retrieshy3 minute 35: 5 retrieshy3 minute 36: 11 retrieshy3 minute 37: 6 retrieshy3 minute 38: 7 retrieshy3 minute 39: 7 retrieshy3 minute 40: 7 retrieshy3 minute 41: 7 retrieshy3 minute 42: 10 retrieshy3 minute 43: 8 retrieshy3 minute 44: 6 retrieshy3 minute 45: 3 retrieshy3 minute 46: 6 retrieshy3 minute 47: 11 retrieshy3 minute 48: 4 retrieshy3 minute 49: 2 retrieshy3 minute 50: 5 retrieshy3 minute 51: 2 retrieshy3 minute 52: 1 retrieshy3 minute 53: 8 retrieshy3 minute 54: 6 retrieshy3 minute 55: 2 retrieshy3 minute 56: 5 retrieshy3 minute 57: 3 retrieshy3 minute 58: 7 retrieshy3 minute 59: 5 retrieshy3 minute 60: 7 retrieshy3 minute 61: 1 retrieshy3 minute 62: 2 retrieshy3 minute 63: 2 retrieshy3 minute 64: 1 retrieshy3 minute 65: 2 retrieshy3 minute 66: 4 retrieshy3 minute 67: 7 retrieshy3 minute 68: 1 retrieshy3 minute 69: 1 retrieshy3 minute 70: 5 retrieshy3 minute 71: 8 retrieshy3 minute 72: 2 retrieshy3 minute 73: 1 retrieshy3 minute 74: 4 retrieshy3 minute 75: 3 retrieshy3 minute 76: 6 retrieshy3 minute 77: 3 retrieshy3 minute 78: 3 retrieshy3 minute 79: 5 retrieshy3 minute 82: 1 retrieshy3 minute 83: 3 retrieshy3 minute 84: 3 retrieshy3 minute 85: 2 retrieshy3 minute 86: 1 retrieshy3 minute 87: 5 retrieshy3 minute 88: 2 retrieshy3 minute 89: 1 retrieshy3 minute 90: 2 retrieshy3 minute 91: 1 retrieshy3 minute 92: 5 retrieshy3 minute 93: 4 retrieshy3 minute 94: 7 retrieshy3 minute 98: 2 retrieshy3 minute 99: 1 retrieshy3 minute 100: 1 retrieshy3 minute 101: 1 retrieshy3 minute 102: 1 retrieshy3 minute 103: 1 retrieshy3 minute 104: 1 retrieshy3 minute 105: 1 retrieshy3 minute 106: 1 retrieshy3 minute 109: 2 retrieshy3 minute 110: 3 retrieshy3 minute 111: 1 retrieshy3 minute 116: 1 retrieshy3 minute 119: 1 retrieshy3 minute 120: 2 retrieshy3 minute 121: 3 retrieshy3 minute 122: 1 retrieshy3 minute 124: 1 retrieshy3 minute 132: 2 retrieshy3 minute 133: 3 retrieshy3 minute 134: 1 retrieshy3 minute 135: 2 retrieshy3 minute 145: 2 retrieshy3 minute 151: 1 retrieshy3 minute 154: 1 retrieshy3 minute 157: 1 retrieshy3 minute 158: 1 retrieshy3 minute 159: 1 retrieshy3 minute 160: 1 retrieshy3 minute 164: 1 retrieshy3 minute 165: 1 retrieshy3 minute 166: 1 retrieshy3 minute 167: 1 retrieshy3 minute 168: 2 retrieshy3 minute 175: 1 retrieshy3 minute 194: 1 retrieshy3591t=0194 min
modelerrorsmedian decision
deepseek-v4-flashremoteprotocol 1 timeout 3,593 unparseable 4149.1s incl. retries
gpt-5.6-lunatimeout 25.4s incl. retries
gemini-3.7-flashtimeout 32.9s incl. retries
hy3connect 1 timeout 59013.9s incl. retries
retry depth — how many times a single decision was re-asked
2,29805001,0001,5002,000attempt 1: 2,298 decisions (deepseek-v4-flash 1,760 gpt-5.6-luna 2 gemini-3.7-flash 3 hy3 533)attempt 12,298attempt 2: 1,144 decisions (deepseek-v4-flash 1,087 hy3 57)attempt 21,144attempt 3: 490 decisions (deepseek-v4-flash 489 hy3 1)attempt 3490attempt 4: 201 decisions (deepseek-v4-flash 201)attempt 4201attempt 5: 56 decisions (deepseek-v4-flash 56)attempt 556attempt 6: 16 decisions (deepseek-v4-flash 16)attempt 616attempt 7: 9 decisions (deepseek-v4-flash 9)attempt 79attempt 8: 7 decisions (deepseek-v4-flash 7)attempt 87deepest chain: attempt 8 x7

Every timeout was retried until the model answered — nobody was substituted. deepseek stalled almost every minute of the 194-minute run (3,635 retries, median decision 49s incl. retries) yet still tied for first; luna won just as many while barely dropping a packet.