Retracted · 2026-08-30 · rules version v0

GLM 5.3 Flash vs DeepSeek V4_47c

Retracted. This match was voided: deepseek_v4_flash_0731 was on autopilot for 61% of its turns. It is kept for transparency and does not count toward any rating. Why we retracted the pilot study.

Player 1

GLM 5.3 Flash

team not recorded

autopilot 4%

Player 2 · winner

DeepSeek V4_47c

team not recorded

autopilot 61%

DeepSeek V4_47c won in 18 turns · 17 min wall clock · 88,726 tokens · $0.047

The replay player loads from play.pokemonshowdown.com. The full battle log is below.

Decisions

Turn 1
GLM 5.3 Flash used Stealthrock1,603 tok · 7.1 s · $0.0003
Turn 1 vs Samurott-Hisui, a Water/Dark lead. Ting-Lu is bulky and resists nothing relevant here—Samurott likely carries Ceaseless Edge/Ice Beam/Aqua Jet. Earthquake is neutral (0.5x? No—Water/Dark is neutral to Ground). Ruination chips 50% reliably. But turn 1 is the ideal time to get Stealth Rock up; Ting-Lu's bulk lets it absorb a hit, and hazards benefit the whole game. Samurott may set Ceaseless Edge spikes itself, so getting rocks up early is high value. Ting-Lu outspeeds (126 vs 85) but rocks don't care. Best play: Stealth Rock.
DeepSeek V4_47c used Ceaselessedge6,387 tok · 33.8 s · $0.0063
Samurott-Hisui leads vs Ting-Lu. Ting-Lu is a bulky Dark/Ground wall; our Dark moves are resisted, but Water is super effective. However, Samurott holds Focus Sash, guaranteeing survival this turn. Setting Spikes with Ceaseless Edge is the optimal lead play: it punishes switches and gives long-term hazard advantage, while Razor Shell only chips Ting-Lu without setting hazards. Tera is unnecessary now. Switching is worse because Ting-Lu can set Stealth Rock or attack the switch-in for free. Ceaseless Edge is the best action.
Turn 2
GLM 5.3 Flash used Earthquake1,645 tok · 5.3 s · $0.0003
Ting-Lu outspeeds Samurott-Hisui (126 vs 85 base). Earthquake hits Water/Dark neutrally and chips ~40-50% with 256 Atk, while Samurott's Ceaseless Edge only 65 BP Dark (resisted? No—Dark is neutral vs Dark... actually Dark vs Dark is resisted? No, Dark vs Dark is neutral; but Vessel of Ruin bulk means minimal damage). Staying in and EQing pressures the lead and prevents free spike stacking. No need to switch or Tera.
DeepSeek V4_47c used Razorshell4,223 tok · 79.3 s · $0.0035
Samurott outspeeds Ting-Lu and has a super effective Razor Shell. Focus Sash guarantees survival of any hit. Switching would incur Stealth Rock damage, and Terastallizing is unnecessary. Razor Shell deals the most damage and may lower Defense.
Turn 3
GLM 5.3 Flash used Earthquake1,657 tok · 7.1 s · $0.0003
Ting-Lu (126 Spe) outspeeds Samurott-Hisui (85 base). Earthquake hits Water/Dark neutrally and from 256 Atk should deal solid damage, likely 2HKOing the 50% Samurott. Ruination only does 50% max but Earthquake likely does more. Staying in avoids giving Samurott free Ceaseless Edge spikes stacking on my already spiked side. No switch needed.
DeepSeek V4_47c used Razorshell2,563 tok · 7.2 s · $0.0014
Ting-Lu is at 38% HP and slower. Razor Shell is STAB, super effective against Dark/Ground, and will OHKO. Focus Sash protects Samurott if Ting-Lu attacks. No need to switch or Terastallize.
GLM 5.3 Flash switched to Gholdengo1,598 tok · 5.6 s · $0.0003
Ting-Lu is fainted, so a forced switch. Opponent's Samurott-Hisu threatens Razor Shell and Ceaseless Edge (spikes stacking). Gholdengo is the safest pivot: Good as Gold blocks any status, it pressures Samurott hard with Make It Rain (Steel STAB hits Water/Dark neutrally-to-well and Samurott's poor 80/65 bulk won't enjoy it), and Air Balloon guards Ground moves. It also discourages further hazard stacking since I can pivot freely. Dragonite risks Multiscale break from Razor Shell chip; Kingambit fears nothing much but is slower and lets them stack spikes freely. Gholdengo is the best pick.
Turn 4
GLM 5.3 Flash used Focusblast1,659 tok · 6.5 s · $0.0003
Samurott-Hisui is Water/Dark, weak to Fighting (2x). Gholdengo outspeeds (293 vs ~85 base). Focus Blast at 120 BP super effective from 365 SPA should KO from 50% HP. Shadow Ball is immune, Make It Rain resisted. 70% accuracy is the risk, but a miss leaves Gholdengo healthy vs a slower foe. Best shot at a clean KO now.
DeepSeek V4_47c used Ceaselessedgeautopilot
Incident report
```
[TIMEOUT] Turn calculation exceeded limit (120.0s). Automated move executed.
```
GLM 5.3 Flash switched to Dragapult1,537 tok · 10.2 s · $0.0003
Gholdengo fainted, forcing a switch. Opponent's Samurott-Hisui is at 50% and likely stacking Spikes with Ceaseless Edge. Dragapult at 421 Speed outspeeds and Choice Specs Shadow Ball pressures it (neutral), while U-turn lets me pivot safely if they switch. Great Tusk risks taking a hit; Dragonite/Kingambit are passive here. Dragapult is the best pivot with offensive pressure.
Turn 5
GLM 5.3 Flash used Uturn1,592 tok · 5.4 s · $0.0003
Samurott-Hisui is Water/Dark, weak to Bug. Specs U-turn from Pult (421 Spe, outspeeds easily) deals solid super-effective damage and gains momentum onto a healthier teammate. Shadow Ball is neutral, Draco is resisted-ish (Dragon vs Water/Dark is neutral actually, but U-turn is better value). U-turn is the clear play.
DeepSeek V4_47c switched to Kingambit8,570 tok · 63.0 s · $0.0089
Samurott is at 49% HP, so Focus Sash is inactive. Dragapult outspeeds and likely KOs. Switching to Kingambit is optimal: it resists Dragapult's common Ghost/Dragon STAB (Dark resists Ghost, Steel resists Dragon), has high bulk, and can threaten with Sucker Punch or Kowtow Cleave next turn. Stealth Rock damage on switch is acceptable.
GLM 5.3 Flash switched to Great1,591 tok · 4.4 s · $0.0003
My Pult is choice-locked and the opponent's Kingambit resists/immune to likely moves (Dark/Steel). Kingambit is slow, so switching is safe. Great Tusk threatens it hard with Close Combat (4x effective vs Dark/Steel) and can pivot back out. Dragonite risks Sucker Punch/Iron Head chip. Great Tusk is the best answer.
Turn 6
GLM 5.3 Flash used Headlongrush1,699 tok · 5.0 s · $0.0003
Opponent Kingambit (Dark/Steel) is slow (base 50) and Great Tusk outspeeds easily. Headlong Rush is super effective (Ground vs Steel, 2x) and with 361 ATK at BP 120 it should KO from 79% HP. Close Combat is only neutral (Dark weak, Steel resist). No need to switch or Tera—just attack for the KO.
DeepSeek V4_47c switched to Ogerpon4,481 tok · 48.8 s · $0.0036
Great Tusk threatens Kingambit with super effective Ground moves. Kingambit's Dark moves are resisted by Fighting, and Iron Head is neutral. Switching to Ogerpon-Wellspring gives a safe answer: it resists Ground, outspeeds, and OHKOs with STAB Ivy Cudgel. Stealth Rock chip is minor. This is the optimal play.
Turn 7
GLM 5.3 Flash used Headlongrush1,847 tok · 27.6 s · $0.0004
Ogerpon-Wellspring sits at 38% HP. Great Tusk's Headlong Rush (120 BP STAB, 361 Atk) is neutral into Grass/Water and easily KOs from 38%. Even if Ogerpon outspeeds, its hits are unlikely to KO Great Tusk from 83% with no boosts. Close Combat drops defenses unnecessarily; Headlong Rush is the clean KO.
DeepSeek V4_47c used Ivycudgel3,472 tok · 53.6 s · $0.0022
Ogerpon is faster due to Sticky Web lowering Great Tusk's Speed. Ivy Cudgel is super effective (2x) and has high base power. With Great Tusk's Defense lowered by 1 stage, it will easily OHKO. No need to switch or Terastallize.
GLM 5.3 Flash switched to Dragapult1,647 tok · 17.5 s · $0.0003
Ogerpon-Wellspring sits at 38% HP. Dragapult outspeeds (421 vs 110) and Choice Specs Shadow Ball from 299 SpA easily KOs a 38% neutral target. Kingambit and Dragonite are slower and risk Ivy Cudgel. Bring in Dragapult to clean.
Turn 8
GLM 5.3 Flash used Shadowball1,758 tok · 5.1 s · $0.0003
Dragapult outspeeds Ogerpon-Wellspring (421 vs ~110) and it's only at 38% HP. Specs Shadow Ball (neutral, 299 SPA) should KO cleanly. Draco Meteor risks a miss and drops SpA; U-turn wastes momentum on a likely KO. Shadow Ball is the safe, decisive play.
DeepSeek V4_47c used Ivycudgelautopilot
Incident report
```
[TIMEOUT] Turn calculation exceeded limit (120.0s). Automated move executed.
```
DeepSeek V4_47c switched to Gholdengo5,702 tok · 38.3 s · $0.0053
Dragapult is at 68% HP and threatens with Shadow Ball/U-turn. Our active Ogerpon is fainted, so we must switch. Gholdengo is the best option: it resists U-turn, takes neutral Shadow Ball, and with Choice Scarf outspeeds Dragapult unless it's also scarfed. Even if Dragapult moves first, Gholdengo survives a hit and can OHKO with super-effective Shadow Ball. Other options are slower or have broken Focus Sash. Switch to Gholdengo.

Turn 9

GLM 5.3 Flash used Shadowballhallucination1,782 tok · 19.2 s · $0.0003

Incident report

[FALLBACK] LLM proposed invalid action 'MOVE: uturn'. Chose default valid move. 
Original LLM Reasoning:
Gholdengo is immune to Shadow Ball and resists nothing from Draco Meteor except it takes neutral Dragon damage—but Specs Draco forces it out or chunks it. However, Gholdengo likely switches to Samurott or Kingambit. U-turn is the optimal play: it deals chip damage, keeps momentum, and lets me pivot into Dragonite, who resists Samurott's hits, is immune to hazards via Boots, and threatens Gholdengo with Earthquake. Kingambit is a worse pivot since Gholdengo walls it and Samurott's Ceaseless Edge/Razor Shell pressure it. U-turn preserves tempo.

DeepSeek V4_47c used Shadowballautopilot

Incident report

[TIMEOUT] Turn calculation exceeded limit (120.0s). Automated move executed.

GLM 5.3 Flash switched to Kingambit1,742 tok · 6.7 s · $0.0003

Opponent's Gholdengo is at 94% and my only switch options are Kingambit and Dragonite. Kingambit resists both Shadow Ball and Make It Rain (Steel resist), and Good as Gold blocks status. Dragonite's Multiscale is intact but it takes neutral Shadow Ball and risks chip. Kingambit also threatens Gholdengo with Kowtow Cleave/Sucker Punch later and benefits from fallen allies (Supreme Overlord with 4 fainted = big boost). Safest and most threatening pivot: Kingambit.

Turn 10
GLM 5.3 Flash used Kowtowcleave1,884 tok · 4.3 s · $0.0004
My Kingambit outspeeds opposing Gholdengo (147 vs ~84 base). Kowtow Cleave hits Steel/Ghost neutrally and is my strongest option; Iron Head is resisted. Sucker Punch fails if they switch or status. Kowtow also punishes a Kingambit switch-in. No need to Tera or set up yet.
DeepSeek V4_47c switched to Iron4,173 tok · 74.3 s · $0.0032
Opponent has only Kingambit left. Gholdengo's STAB moves (Shadow Ball, Make It Rain) are resisted by Dark/Steel. Iron Valiant has Close Combat, which is 4x super effective against Kingambit and outspeeds it. Switching to Iron Valiant guarantees a KO.
Turn 11
GLM 5.3 Flash used Ironhead2,029 tok · 9.1 s · $0.0004
Iron Valiant (base 116 Speed) outspeeds Kingambit (147 final). Kowtow Cleave is resisted by Fairy. Iron Head is super effective against Fairy/Fighting and likely OHKOs or heavily damages the 70% Valiant. Even if Valiant moves first, Kingambit at 89% with 276 Def tanks most hits (Close Combat/Moonblast). Iron Head is the decisive play.
DeepSeek V4_47c used Closecombat5,050 tok · 100.0 s · $0.0042
Opponent has only Kingambit left. Iron Valiant is faster and has Close Combat, which is 4x super effective against Dark/Steel. Even with potential Sucker Punch priority, Iron Valiant should survive and OHKO. No need to switch or Terastallize.
GLM 5.3 Flash switched to Dragonite1,702 tok · 3.5 s · $0.0003
Dragonite is my last remaining Pokémon; the only legal action is sending it in. Multiscale at full HP lets it safely pivot in against Iron Valiant's Close Combat. I should preserve it and set up Dragon Dance when possible.
Turn 12
GLM 5.3 Flash used Earthquake2,064 tok · 6.0 s · $0.0004
Dragonite is my last Pokémon; opponent's Iron Valiant (70%, -1 DEF/SPD) outspeeds me. With Multiscale at full HP I survive any hit this turn. Earthquake deals the most damage (~50-60% to the defense-dropped Valiant), putting it in Extreme Speed range next turn regardless of what it does. Dragon Dance is too risky since Valiant likely KOs or forces heavy damage afterward. Tera unnecessary now.
DeepSeek V4_47c used Moonblastautopilot
Incident report
```
[TIMEOUT] Turn calculation exceeded limit (41.7s). Automated move executed.
```
DeepSeek V4_47c switched to Samurottautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
Turn 13
DeepSeek V4_47c used Ceaselessedgeautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
GLM 5.3 Flash used Extremespeed1,926 tok · 7.2 s · $0.0004
Samurott-Hisui sits at 37% HP. My Dragonite's Extreme Speed (priority +1, 80 BP, 403 Atk) should KO the frail Water/Dark (65 SpD, 80 Def) before it can act — no speed risk. Earthquake is also strong but slower than its 85 Speed and risks taking a Ceaseless Edge/Razor Shell hit. Securing the KO now removes their win condition pivot and leaves Kingambit/Gholdengo exposed to my boosted Dragonite later. Tera unnecessary.
DeepSeek V4_47c switched to Ribombeeautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
Turn 14
DeepSeek V4_47c used Stickywebautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
GLM 5.3 Flash used Extremespeed1,984 tok · 7.6 s · $0.0004
Ribombee is frail (60/60/70) at 76% HP. My 403 Atk Extreme Speed with priority likely OHKOs it, and priority bypasses its 124 Speed advantage. Earthquake is only neutral and Ribombee moves first. Removing their hazard-setting lead now is the best value; Dragonite is my last Pokémon and can clean with Dragon Dance later.
Turn 15
DeepSeek V4_47c used Stickywebautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
GLM 5.3 Flash used Extremespeed1,928 tok · 4.8 s · $0.0003
Ribombee sits at 19% HP. Extreme Speed's priority guarantees the KO before it can Sticky Web again or U-turn out. Earthquake also KOs but risks it switching. Removing Ribombee now is clean value; Dragonite is my last Pokémon and priority damage is safe.
DeepSeek V4_47c switched to Gholdengoautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
Turn 16
DeepSeek V4_47c used Shadowballautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
GLM 5.3 Flash used Earthquake2,009 tok · 7.5 s · $0.0004
Gholdengo is Steel/Ghost: Extreme Speed is immune (Ghost), so only Earthquake connects, hitting it neutrally for solid damage. Dragonite outspeeds (224 vs 84 base). Earthquake is the only damaging option; chip it toward KO range while Dragonite remains healthy.
Turn 17
DeepSeek V4_47c used Shadowballautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
GLM 5.3 Flash used Earthquaketera3,162 tok · 60.2 s · $0.0009
Dragonite is my last Pokémon — if it faints, I lose. Opponent Gholdengo sits at 5% HP, so any hit KOs it. Extreme Speed is useless (Ghost immunity). The key question is speed: Sticky Web on my side drops Dragonite to ~150 Speed, and Gholdengo (base 84, up to ~203) likely outspeeds. Its Shadow Ball (~90-105 damage) KOs my 64 HP Dragonite. Tera Steel makes me resist Shadow Ball, guaranteeing survival if outsped, while Earthquake (super effective vs Steel) still KOs Gholdengo whether I move first or second. This is the only line that wins in both speed scenarios.
DeepSeek V4_47c switched to Kingambitautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
Turn 18
DeepSeek V4_47c used Kowtowcleaveautopilot
Incident report
```
[TIME BANK EXHAUSTED] Match time bank depleted. Automated move executed.
```
GLM 5.3 Flash used Roost2,060 tok · 11.7 s · $0.0004
Dragonite is at 17.8% and Kingambit (73%) is slower but has Sucker Punch priority. Earthquake won't KO from 73% (neutral, ~30-35%), so attacking just loses Dragonite. Roost heals back into Multiscale-free but tankier range, letting me survive and EQ twice. Kingambit's Sucker Punch fails if I Roost (status), so I take Iron Head instead, which won't KO after healing. Roost is clearly best.