Benchmark · Champions BSS Reg M-B · own teams · 1,219 rated games · updated 2026-10-17

Rotom Champ Bench

Pokémon Champions singles, Regulation M-B: level 50, bring six and pick three, the team each model built.

Pokémon Champions singles under Regulation M-B, the rules of the game's ranked Battle Stadium Singles. Each model builds six Pokémon from the regulation's list, with Champions' Stat Points (at most 32 in one stat, 66 in all), a different item on each and Mega Evolution instead of Terastallization, then brings three of the six to each battle. Everyone fights at level 50, and games are short.

The teams

Ratings are on the Elo scale, fitted by Bradley-Terry over this board's rated games. Games are only played once; each leaderboard is a different selection of them. GPT-OSS 20B is fixed at 1500 by definition; every other number is relative to it.3Models with fewer than 16 games are marked provisional. Click a column to sort.

StructurePrice
#ModelRating (95% CI)1GamesWLWin rateHalluc.2Tokens / turnCost / game
1Gemini 4 Argon2340+213−3551817194%0.0%1,943$0.240
2GPT-6.1 Sol2159+151−3341613381%0.0%212$0.091
3Gemini 3.8 Flash2009+216−2242117481%5.8%286$0.030
4Muse Wonder1854+273−27198189%0.0%926$0.227
5MiMo V2.6 Flash1790+131−11579611877%2.4%855$0.006
6Gemma 4 31B1776+125−11281592273%3.3%754$0.005
7DeepSeek V4.1 Flash1773+139−12260461477%2.2%1,109$0.012
8GPT-6 Luna Decisions1766+137−11478591976%0.0%0$0.003
9Muse Spark Open1761+152−15830171357%0.2%506$0.070
10Muse Spark 1.31746+116−11781592273%0.1%1,590$0.008
11Claude Opus 5.51738+155−13730201067%0.0%714$0.266
12DeepSeek V4 Pro1726+366−24997278%0.0%2,797$0.107
13Grok 4.71726+363−24297278%0.0%3,783$0.304
14Kimi K31726+363−26697278%0.0%204$0.069
15MiMo V2.6 Pro1726+360−26397278%2.2%2,155$0.026
—PokéChamp1714+95−100141835859%———
16GPT-6 Luna1695+126−10981552668%0.0%369$0.005
17Nemotron 3 Ultra1688+385−29253260%20.3%1,300free
18Claude Sonnet 5.51684+157−15730181260%0.0%447$0.082
19Claude Haiku 5.51682+123−10184552966%1.3%926$0.009
20GPT-6 Luna Pro1652+173−1782112957%0.0%1,380$0.016
21Solar Pro 41648+131−11860382263%0.7%1,781$0.009
22Mercury Decide1626+111−10190543660%0.0%0$0.001
23MiniMax M31621+269−23396367%3.2%6,037$0.108
24GLM 5.3 Flash1618+121−11081483359%0.5%201$0.004
25GLM 5.31618+201−205159660%0.0%244$0.015
26Bonsai 2 27B1601+114−10490513957%1.3%1,784free
27Qwen3.8 Flash1570+125−9775433257%0.0%1,762$0.014
28Qwen3.8 27B1544+148−15230141647%0.0%1,606$0.031
29Qwen3.8 Max Prime1523+244−23795456%0.0%1,203$0.254
30Ling 3.1 Flash1506+265−28673443%26.0%9,210$0.030
31Jev 1.131503+118−11175363948%0.0%0$0.002
32GPT-OSS 20B1500131518039%3.6%1,279free
33Nemotron 3.5 Super VL1470+126−11269323746%0.0%3,406free
34Ornith 1.5 35B1443+122−10975314441%4.0%2,164free
35Microsoft-Decision-11430+126−11775304540%0.0%0$0.001
36Drex v2.01391+125−10775274836%0.0%0$0.003
37Perplexity Decider V1.1 27B1391+139−11075274836%0.0%0$0.001
38Ling 3.0 Tiny1316+131−11069214830%2.8%2,352free
39Drex v1.51300+135−12269204929%10.0%0free
40MiniCPM5 2B1276+127−12075195625%3.4%1,607free
41Laya RotomArena1180+311−33592722%0.0%0free
—Chimps at the Zoo1090+130−1481241211210%———
42Laya1024+165−186757689%2.0%0free
43Solar Decide Flash877+198−313753724%0.1%0$0.002
Rules

Every game records the rules it was played under; a board only rates games that match their version's stored rules.

VersionFormatTeamsBringThinking per turnTokens per turnPrompt
c5gen9championsrotomarenabssregmbown (drafted)3 of 64,096 tokens8,1928978284d4c8b · draft 09e628a45602
400600800100012001400160018002000220024002600Gemini 4 ArgonGPT-6.1 SolGemini 3.8 FlashMuse WonderMiMo V2.6 FlashGemma 4 31BDeepSeek V4.1 FlashGPT-6 Luna DecisionsMuse Spark OpenMuse Spark 1.3Claude Opus 5.5DeepSeek V4 ProGrok 4.7Kimi K3MiMo V2.6 ProPokéChampGPT-6 LunaNemotron 3 UltraClaude Sonnet 5.5Claude Haiku 5.5GPT-6 Luna ProSolar Pro 4Mercury DecideMiniMax M3GLM 5.3 FlashGLM 5.3Bonsai 2 27BQwen3.8 FlashQwen3.8 27BQwen3.8 Max PrimeLing 3.1 FlashJev 1.13GPT-OSS 20BNemotron 3.5 Super VLOrnith 1.5 35BMicrosoft-Decision-1Drex v2.0Perplexity Decider V1.1 27BLing 3.0 TinyDrex v1.5MiniCPM5 2BLaya RotomArenaChimps at the ZooLayaSolar Decide Flashmodelbotanchor (fixed at 1500)
Figure 1. Ratings with 95% parametric bootstrap intervals; the bars end at the interval bounds. The hollow point is the anchor, whose interval is zero by construction, and the vertical line marks its rating. Overlapping intervals mean we cannot tell two models apart, which is most of them.
Most attractive quadrant800900100011001200130014001500160017001800190020002100220023002400$0.0003$0.001$0.003$0.01$0.03$0.1$0.3$1freeCost per game (USD, log scale)RatingPokéChamp 1714Chimps at the Zoo 1090Gemini 4 Argon: 2340, $0.240 a gameGemini 4 ArgonGPT-6.1 Sol: 2159, $0.091 a gameGPT-6.1 SolGemini 3.8 Flash: 2009, $0.052 a gameGemini 3.8 FlashMuse Wonder: 1854, $0.227 a gameMuse WonderMiMo V2.6 Flash: 1790, $0.006 a gameMiMo V2.6 FlashGemma 4 31B: 1776, $0.015 a gameDeepSeek V4.1 Flash: 1773, $0.021 a gameGPT-6 Luna Decisions: 1766, $0.003 a gameGPT-6 Luna DecisionsMuse Spark Open: 1761, $0.070 a gameMuse Spark OpenMuse Spark 1.3: 1746, $0.009 a gameClaude Opus 5.5: 1738, $0.276 a gameDeepSeek V4 Pro: 1726, $0.107 a gameGrok 4.7: 1726, $0.304 a gameGrok 4.7Kimi K3: 1726, $0.069 a gameKimi K3MiMo V2.6 Pro: 1726, $0.026 a gameGPT-6 Luna: 1695, $0.007 a gameGPT-6 LunaNemotron 3 Ultra: 1688, free a gameNemotron 3 UltraClaude Sonnet 5.5: 1684, $0.079 a gameClaude Haiku 5.5: 1682, $0.009 a gameGPT-6 Luna Pro: 1652, $0.022 a gameSolar Pro 4: 1648, $0.012 a gameMercury Decide: 1626, $0.001 a gameMercury DecideMiniMax M3: 1621, $0.108 a gameMiniMax M3GLM 5.3 Flash: 1618, $0.004 a gameGLM 5.3 FlashGLM 5.3: 1618, $0.015 a gameBonsai 2 27B: 1601, free a gameQwen3.8 Flash: 1570, $0.014 a gameQwen3.8 FlashQwen3.8 27B: 1544, $0.032 a gameQwen3.8 27BQwen3.8 Max Prime: 1523, $0.254 a gameQwen3.8 Max PrimeLing 3.1 Flash: 1506, $0.030 a gameLing 3.1 FlashJev 1.13: 1503, $0.002 a gameJev 1.13GPT-OSS 20B: 1500, free a gameGPT-OSS 20BNemotron 3.5 Super VL: 1470, free a gameOrnith 1.5 35B: 1443, free a gameMicrosoft-Decision-1: 1430, $0.001 a gameDrex v2.0: 1391, $0.003 a gameDrex v2.0Perplexity Decider V1.1 27B: 1391, $0.001 a gameLing 3.0 Tiny: 1316, free a gameLing 3.0 TinyDrex v1.5: 1300, free a gameMiniCPM5 2B: 1276, free a gameLaya RotomArena: 1180, free a gameLaya RotomArenaLaya: 1024, free a gameLayaSolar Decide Flash: 877, $0.002 a gameSolar Decide Flash
Figure 2. Rating against cost per game. The dotted line is the Pareto frontier: no model is both cheaper and better than a model on it. The shaded quadrant is better and cheaper than the median paid model. The dashed lines are the control bots, who cost nothing and are not invited to the frontier.

Head to head

Wins–losses of the row model against the column model. Click a record for its games and replays.

Gemini 4 ArgonGPT-6.1 SolGemini 3.8 FlashMuse WonderMiMo V2.6 FlashGemma 4 31BDeepSeek V4.1 FlashGPT-6 Luna DecisionsMuse Spark OpenMuse Spark 1.3Claude Opus 5.5DeepSeek V4 ProGrok 4.7Kimi K3MiMo V2.6 ProPokéChampGPT-6 LunaNemotron 3 UltraClaude Sonnet 5.5Claude Haiku 5.5GPT-6 Luna ProSolar Pro 4Mercury DecideMiniMax M3GLM 5.3 FlashGLM 5.3Bonsai 2 27BQwen3.8 FlashQwen3.8 27BQwen3.8 Max PrimeLing 3.1 FlashJev 1.13GPT-OSS 20BNemotron 3.5 Super VLOrnith 1.5 35BMicrosoft-Decision-1Drex v2.0Perplexity Decider V1.1 27BLing 3.0 TinyDrex v1.5MiniCPM5 2BLaya RotomArenaChimps at the ZooLayaSolar Decide Flash
Gemini 4 Argon·······································
GPT-6.1 Sol········································
Gemini 3.8 Flash·····································
Muse Wonder·········································
MiMo V2.6 Flash·················
Gemma 4 31B···················
DeepSeek V4.1 Flash························
GPT-6 Luna Decisions··················
Muse Spark Open···································
Muse Spark 1.3·················
Claude Opus 5.5···································
DeepSeek V4 Pro·········································
Grok 4.7·········································
Kimi K3·········································
MiMo V2.6 Pro·········································
PokéChamp·
GPT-6 Luna·················
Nemotron 3 Ultra··········································
Claude Sonnet 5.5····································
Claude Haiku 5.5················
GPT-6 Luna Pro······································
Solar Pro 4························
Mercury Decide················
MiniMax M3·········································
GLM 5.3 Flash·················
GLM 5.3·······································
Bonsai 2 27B················
Qwen3.8 Flash····················
Qwen3.8 27B·····································
Qwen3.8 Max Prime·········································
Ling 3.1 Flash·········································
Jev 1.13···················
GPT-OSS 20B
Nemotron 3.5 Super VL·····················
Ornith 1.5 35B···················
Microsoft-Decision-1···················
Drex v2.0···················
Perplexity Decider V1.1 27B···················
Ling 3.0 Tiny·····················
Drex v1.5·····················
MiniCPM5 2B···················
Laya RotomArena·········································
Chimps at the Zoo··
Laya···················
Solar Decide Flash···················

    1. 95% intervals come from 1,000 parametric bootstrap refits. When an interval covers half the chart, the model has not played much yet, or plays very inconsistently, or both.
    2. Hallucination rate: the share of turns where the model gave no usable answer (named a move it does not have, answered in a format we could not read, rambled past its token budget, timed out or errored) and the default move was played on its behalf. Matches where either side does this on more than 50% of its turns are voided.
    3. Ratings need a reference point. We chose a local open-weights model because it is always available, costs nothing to run, and cannot be deprecated by its vendor while we sleep.
    4. Older versions of this benchmark, retired when their rules changed, are in the archive.