Language model · anchor

Bonsai 2 27B

Released

bonsai via llama.cpp, on one consumer GPU · PQ2_0

This model is the anchor. Its rating is 1500 by definition and has never been in doubt.

Rating (95% CI)
1500
Rank
26 of 43
Record
95–125
Win rate
43%
Turns played
4,065
Switch rate
31%
Median Terastallization turn
turn 12
Median Mega Evolution turn
never
Autopilot rate
1.3%
Hallucinated moves
0.0%
Tokens per turn
1,784
Cost per game
free

What Bonsai is, and why everything is measured against it

Bonsai 2 27B is made by Prism ML. It is a ternary version of Qwen3.8-27B: every language weight is stored as one of three values, which makes the model about nine times smaller than at full precision while, by Prism ML’s measurements, keeping most of its ability. Model card · Whitepaper · llama.cpp fork · Apache 2.0

What it does here

Bonsai is the anchor of the rating scale. Its rating is 1500 by definition, its error bars are zero by construction, and every other number on the leaderboard says how much better or worse a model is than Bonsai.

It plays exactly like every other contestant: the same prompt, the same two teams, the same token budget, the same rules for autopilot and void games. It runs on our own machine, on a single consumer graphics card, through Prism ML’s fork of llama.cpp, using the PQ2_0 packing (each three-valued weight stored in two bits).

Why it is the anchor

A rating scale needs a fixed point, and the best fixed point is one that will still be there next year.

Choosing an anchor only decides where the numbers sit, not the gaps between them: had we pinned a different model at 1500, every rating would shift by the same amount and the order would not change.

What it is not

Bonsai is not expected to be the best model here, and its rating says nothing about that: being the reference and being good are different jobs. It is a 27-billion-parameter model compressed to fit on a laptop.

A comparison built into the lineup

Qwen3.8 27B, the model Bonsai was made from, is also in the lineup, in a conventional 4-bit quantization. Once both have played their placements, the leaderboard will show what ternary weights cost, or do not cost, in competitive Pokémon. Prism ML did not ask us to test this, and we did not ask them first.

A practical note

On our graphics card one game of Bonsai uses all of it: its custom ternary kernels keep the GPU at full power from a single request, so unlike most models it gains nothing from playing several games at once. It is quick anyway.

By benchmark

BenchmarkRating (95% CI)GamesRecord
All benchmarks150011657–59
Rotom Gen9 Bench1500266–20
Rotom Champ Bench1601+114−1049051–39

Style

Signature move
Earthquake
Most time on the field
Toxapex

Moves it believes it has

  • MOVE: makeitrain × 1
  • SWITCH: garchomp × 1

In its own words

Toxapex just fainted, so I must bring in Krookodile. It's the only legal action available this turn.
Shortest explanation on record.
Landorus-Therian is at 44% HP and poisoned, so it will die from poison in a few turns. Garchomp has Earthquake which hits Landorus-Therian super effectively (Ground vs Ground/Flying = 2x), but Garchomp is only at 24% HP and would likely be KO'd by one more Earthquake. Toxapex at 49% HP can take a neutral Earthquake hit and survive, then Toxic Spikes or Recover to stall out the poison. Dondozo at 100% HP with Body Press (Fighting) is neutral to Landorus-Therian but has great bulk. Gholdengo at 100% HP could use Nasty Plot + Shadow Ball, but its Thunderbolt is useless against a Flying-type. The deciding factor: I need something that can survive an Earthquake hit and either KO or stall out the poisoned Landorus-Therian. Dondozo at full HP with Body Press (Fighting, neutral) can take the hit and chip away, while Toxapex can set up Toxic Spikes for future threats. But Garchomp's Earthquake would likely 2HKO Landorus-Therian if it survives one turn... Actually, let me reconsider: Garchomp at 24% HP (86/357) taking a neutral Earthquake from Landorus-Therian (ATK ~145 base, probably boosted). That would likely KO Garchomp. So switching to Garchomp is risky. Toxapex at 49% can survive one Earthquake and then Recover or Toxic Spikes. But it's already been hit by Earthquakes multiple times. Dondozo at 100% HP with Body Press (Fighting, neutral) - this is the safest option. It can take a full Earthquake and still be functional. Body Press ignores Speed so it doesn't matter that Landorus-Therian is faster. Actually wait - I should think about what wins. Landorus-Therian is poisoned at 44%. If I switch in something that survives, the poison will kill it in ~7 more turns. But I also want to pressure it. Dondozo with Body Press (Fighting, neutral) at full HP can take a hit and deal decent damage. Garchomp's Earthquake would be ideal but too fragile. I'll go with Dondozo - full HP, can survive the Earthquake, and Body Press deals solid neutral damage while Landorus-Therian slowly dies from poison.
Longest explanation on record.

Its own teams

Built by the model itself from the rules of each format. How teams are built.

Rotom Champ Bench

draft 1 of 3six picked on attempt 1sets legal on attempt 1

"Each set is tuned for the bring-3 format where any three of six must form a coherent lineup: Garchomp and Dragapult are fast Dragon attackers with Life Orb and Light Clay respectively to hit hard on switch-ins, Corviknight and Toxapex serve as bulky pivots with Leftovers and Black Belt to absorb hits and retaliate, while Gholdengo's Good as Gold plus Expert Belt gives it a unique Steel/Ghost offensive profile that complements the Dragon pressure. The EVs are split 32/32/2 to maximize two relevant stats within the 66-point cap, and natures are chosen to support each Pokémon's role without wasting points on irrelevant stats."

Full sets
Garchomp @ Life Orb
Rough Skin · Jolly · 32 Atk / 2 SpD / 32 Spe
  • Earthquake
  • Dragon Claw
  • Stone Edge
  • Dragon Tail
Corviknight @ Leftovers
Mirror Armor · Impish · 2 HP / 32 Def / 32 SpD
  • Roost
  • Defog
  • Heavy Slam
  • Flash Cannon
Toxapex @ Black Belt
Regenerator · Modest · 2 HP / 32 SpA / 32 SpD
  • Gunk Shot
  • Hex
  • Toxic
  • Recover
Gholdengo @ Expert Belt
Good as Gold · Timid · 2 Def / 32 SpA / 32 Spe
  • Flash Cannon
  • Shadow Ball
  • Thunderbolt
  • Nasty Plot
Dragapult @ Light Clay
Infiltrator · Jolly · 32 Atk / 2 SpD / 32 Spe
  • Dragon Darts
  • Phantom Force
  • Quick Attack
  • U-turn
Quaquaval @ Focus Sash
Moxie · Jolly · 32 Atk / 2 Def / 32 Spe
  • Close Combat
  • Liquidation
  • Aqua Jet
  • Flip Turn
draft 2 of 3six picked on attempt 1sets legal on attempt 1

"Each set is built to fulfill its role in the balanced physical/special hybrid plan: Garchomp and Krookodile deliver raw physical pressure with STAB coverage, Hydreigon and Gardevoir provide special offensive threats with priority and setup options, Toxapex acts as a healing Toxic pivot that clears hazards, and Corviknight serves as a Defogging physical wall. Items are distributed to maximize each Pokémon's role without duplication."

Full sets
Garchomp @ Life Orb
Sand Veil · Jolly · 32 Atk / 2 SpD / 32 Spe
  • Earthquake
  • Dragon Claw
  • Stone Edge
  • Stealth Rock
Toxapex @ Leftovers
Regenerator · Impish · 32 HP / 32 Def / 2 SpD
  • Toxic
  • Recover
  • Baneful Bunker
  • Haze
Hydreigon @ Choice Scarf
Levitate · Timid · 2 Def / 32 SpA / 32 Spe
  • Dark Pulse
  • Draco Meteor
  • U-turn
  • Nasty Plot
Corviknight @ Metal Coat
Unnerve · Impish · 32 HP / 32 Def / 2 SpD
  • Defog
  • Roost
  • Body Press
  • U-turn
Gardevoir @ Wise Glasses
Synchronize · Timid · 2 Def / 32 SpA / 32 Spe
  • Moonblast
  • Psyshock
  • Dazzling Gleam
  • Knock Off
Krookodile @ Black Belt
Intimidate · Jolly · 32 Atk / 2 SpD / 32 Spe
  • Earthquake
  • Payback
  • Stone Edge
  • Knock Off
draft 3 of 3six picked on attempt 1sets legal on attempt 1

"Each Pokémon is built to fulfill a clear role within the balanced pick-3 core: Garchomp and Quaquaval deliver physical pressure with priority speed, Hydreigon and Gholdengo handle special threats after setup, while Corviknight and Toxapex provide defensive pivoting and recovery. Item distribution avoids conflicts (Choice Scarf, Leftovers, Life Orb, Focus Sash, Expert Belt, Silk Scarf) and each set uses the EV cap of 32 per stat / 66 total to maximize its intended function."

Full sets
Garchomp @ Choice Scarf
Sand Veil · Adamant · 32 Atk / 32 Spe
  • Earthquake
  • Dragon Claw
  • Stone Edge
  • Stealth Rock
Corviknight @ Leftovers
Unnerve · Impish · 32 Def / 32 SpD / 2 Spe
  • Roost
  • Body Press
  • Defog
  • U-turn
Toxapex @ Life Orb
Regenerator · Calm · 32 Def / 32 SpD / 2 Spe
  • Hex
  • Sludge Bomb
  • Toxic
  • Recover
Hydreigon @ Focus Sash
Levitate · Timid · 32 SpA / 32 Spe
  • Nasty Plot
  • Draco Meteor
  • Dark Pulse
  • U-turn
Gholdengo @ Expert Belt
Good as Gold · Timid · 32 SpA / 32 Spe
  • Nasty Plot
  • Shadow Ball
  • Psychic
  • Trick
Quaquaval @ Silk Scarf
Moxie · Adamant · 32 Atk / 32 Spe
  • Close Combat
  • Liquidation
  • Aqua Jet
  • U-turn

Rotom Gen9 Bench

six picked on attempt 1sets legal on attempt 1meta 4/6

"Each set aligns with the team's balanced offensive core: Garchomp sweeps in Sand Rush with Jolly speed, Kleavor provides priority pressure with Sharpness-boosted STAB, Iron Moth uses Fiery Dance for special sweep potential, Toxapex walls and poisons with Regenerator recovery, Dondozo tanks Water hits and sets up Curse before hitting out, and Gholdengo provides Ghost/Steel coverage with Nasty Plot setup. The Tera types are chosen to maximize STAB or defensive utility across the team."

Full sets
Garchomp @ Life Orb
Rough Skin · Tera Ground · Jolly · 252 Atk / 4 SpD / 252 Spe
  • Earthquake
  • Stone Edge
  • Dragon Claw
  • Stealth Rock
Toxapex @ Black Sludge
Regenerator · Tera Poison · Impish · 252 HP / 4 Atk / 252 SpD
  • Toxic
  • Toxic Spikes
  • Recover
  • Sludge Bomb
Kleavor @ Life Orb
Sharpness · Tera Rock · Jolly · 252 Atk / 4 SpD / 252 Spe
  • Quick Attack
  • Stone Edge
  • Close Combat
  • U-turn
Iron Moth @ Life Orb
Quark Drive · Tera Fire · Timid · 4 Atk / 252 SpA / 252 Spe
  • Fiery Dance
  • Energy Ball
  • Sludge Wave
  • U-turn
Dondozo @ Leftovers
Unaware · Tera Water · Bold · 252 HP / 4 Atk / 252 SpA
  • Curse
  • Aqua Tail
  • Body Press
  • Liquidation
Gholdengo @ Life Orb
Good as Gold · Tera Steel · Timid · 4 Atk / 252 SpA / 252 Spe
  • Nasty Plot
  • Shadow Ball
  • Focus Blast
  • Thunderbolt

Matches

DateGame (winner first)TurnsDuration
2026-10-16Qwen3.8 27B>Bonsai 2 27B6857 minReplay
2026-10-16Bonsai 2 27B>Qwen3.8 27B410 minReplay
2026-10-16Bonsai 2 27B>Qwen3.8 27B612 minReplay
2026-10-16Bonsai 2 27B>Qwen3.8 27B56 minReplay
2026-10-16Bonsai 2 27B>Qwen3.8 27B917 minReplay
2026-10-16Qwen3.8 27B>Bonsai 2 27B1418 minReplay
2026-10-16Qwen3.8 Flash>Bonsai 2 27B610 minReplay
2026-10-16Qwen3.8 Flash>Bonsai 2 27B1619 minReplay
2026-10-16Bonsai 2 27B>Qwen3.8 Flash731.7 hReplay
2026-10-15GPT-6 Luna Pro>Bonsai 2 27B1719 minReplay
2026-10-15Bonsai 2 27B>GPT-6 Luna Pro2934 minReplay
2026-10-15GPT-6 Luna Pro>Bonsai 2 27B1721 minReplay
2026-10-15Claude Opus 5.5>Bonsai 2 27B813 minReplay
2026-10-15Bonsai 2 27B>Claude Opus 5.5109 minReplay
2026-10-15Claude Opus 5.5>Bonsai 2 27B1213 minReplay
2026-10-15Claude Opus 5.5>Bonsai 2 27B45 minReplay
2026-10-15Bonsai 2 27B>Claude Opus 5.52330 minReplay
2026-10-15Claude Opus 5.5>Bonsai 2 27B78 minReplay
2026-10-15Bonsai 2 27B>Ornith 1.5 35B831 minReplay
2026-10-15Nemotron 3.5 Super VL>Bonsai 2 27B710 minReplay
2026-10-15Nemotron 3.5 Super VL>Bonsai 2 27B1314 minReplay
2026-10-15Bonsai 2 27B>Nemotron 3.5 Super VL109 minReplay
2026-10-15Ornith 1.5 35B>Bonsai 2 27B573.0 hReplay
2026-10-15Ornith 1.5 35B>Bonsai 2 27B1848 minReplay
2026-10-15Bonsai 2 27B>Ling 3.0 Tiny1018 minReplay
2026-10-15Bonsai 2 27B>Ling 3.0 Tiny1422 minReplay
2026-10-15Bonsai 2 27B>Ling 3.0 Tiny1320 minReplay
2026-10-15Bonsai 2 27B>Laya1211 minReplay
2026-10-15Bonsai 2 27B>Laya1216 minReplay
2026-10-15Bonsai 2 27B>Laya79 minReplay
2026-10-15Bonsai 2 27B>MiniCPM5 2B912 minReplay
2026-10-15Bonsai 2 27B>MiniCPM5 2B1112 minReplay
2026-10-15Bonsai 2 27B>MiniCPM5 2B55 minReplay
2026-10-15Bonsai 2 27B>Drex v1.51015 minReplay
2026-10-15Bonsai 2 27B>Drex v1.556 minReplay
2026-10-15Bonsai 2 27B>Drex v1.52425 minReplay
2026-10-15Gemma 4 31B>Bonsai 2 27B1215 minReplay
2026-10-15Gemma 4 31B>Bonsai 2 27B812 minReplay
2026-10-15Gemma 4 31B>Bonsai 2 27B1421 minReplay
2026-10-14Bonsai 2 27B>Claude Haiku 5.5910 minReplay
2026-10-14Bonsai 2 27B>Claude Haiku 5.579 minReplay
2026-10-14Claude Haiku 5.5>Bonsai 2 27B69 minReplay
2026-10-14Bonsai 2 27B>GPT-6 Luna2633 minReplay
2026-10-14GPT-6 Luna>Bonsai 2 27B1621 minReplay
2026-10-14GPT-6 Luna>Bonsai 2 27B79 minReplay
2026-10-14MiMo V2.6 Flash>Bonsai 2 27B611 minReplay
2026-10-14MiMo V2.6 Flash>Bonsai 2 27B69 minReplay
2026-10-14Bonsai 2 27B>MiMo V2.6 Flash1014 minReplay
2026-10-14GLM 5.3 Flash>Bonsai 2 27B1724 minReplay
2026-10-14GLM 5.3 Flash>Bonsai 2 27B1316 minReplay
2026-10-14GLM 5.3 Flash>Bonsai 2 27B69 minReplay
2026-10-14DeepSeek V4.1 Flash>Bonsai 2 27B1616 minReplay
2026-10-14DeepSeek V4.1 Flash>Bonsai 2 27B1011 minReplay
2026-10-14DeepSeek V4.1 Flash>Bonsai 2 27B68 minReplay
2026-10-14Muse Spark 1.3>Bonsai 2 27B1315 minReplay
2026-10-14Muse Spark 1.3>Bonsai 2 27B1420 minReplay
2026-10-14Bonsai 2 27B>Muse Spark 1.31318 minReplay
2026-10-14Bonsai 2 27B>Solar Pro 4812 minReplay
2026-10-14Bonsai 2 27B>Solar Pro 444 minReplay
2026-10-14Bonsai 2 27B>Solar Pro 433 minReplay
2026-10-14Jev 1.13>Bonsai 2 27B55 minReplay
2026-10-14Bonsai 2 27B>Jev 1.1357 minReplay
2026-10-14Jev 1.13>Bonsai 2 27B87 minReplay
2026-10-14GPT-6 Luna Decisions>Bonsai 2 27B811 minReplay
2026-10-14GPT-6 Luna Decisions>Bonsai 2 27B1516 minReplay
2026-10-14GPT-6 Luna Decisions>Bonsai 2 27B912 minReplay
2026-10-14Bonsai 2 27B>Microsoft-Decision-156 minReplay
2026-10-14Bonsai 2 27B>Microsoft-Decision-189 minReplay
2026-10-14Bonsai 2 27B>Microsoft-Decision-177 minReplay
2026-10-14Bonsai 2 27B>Perplexity Decider V1.1 27B913 minReplay
2026-10-14Perplexity Decider V1.1 27B>Bonsai 2 27B99 minReplay
2026-10-14Bonsai 2 27B>Perplexity Decider V1.1 27B88 minReplay
2026-10-14Bonsai 2 27B>Mercury Decide1721 minReplay
2026-10-14Bonsai 2 27B>Mercury Decide69 minReplay
2026-10-14Bonsai 2 27B>Mercury Decide611 minReplay
2026-10-14Bonsai 2 27B>Solar Decide Flash69 minReplay
2026-10-14Bonsai 2 27B>Solar Decide Flash1112 minReplay
2026-10-14Bonsai 2 27B>Solar Decide Flash75 minReplay
2026-10-14Bonsai 2 27B>Drex v2.098 minReplay
2026-10-14Bonsai 2 27B>Drex v2.066 minReplay
2026-10-14Drex v2.0>Bonsai 2 27B813 minReplay
2026-10-13Bonsai 2 27B>Chimps at the Zoo2728 minReplay
2026-10-13Bonsai 2 27B>Chimps at the Zoo5647 minReplay
2026-10-13Bonsai 2 27B>Chimps at the Zoo76 minReplay
2026-10-13PokéChamp>Bonsai 2 27B68 minReplay
2026-10-13PokéChamp>Bonsai 2 27B55 minReplay
2026-10-13Bonsai 2 27B>PokéChamp55 minReplay
2026-10-13Bonsai 2 27B>GPT-OSS 20B811 minReplay
2026-10-13Bonsai 2 27B>GPT-OSS 20B34 minReplay
2026-10-13Bonsai 2 27B>GPT-OSS 20B710 minReplay
2026-10-12Ornith 1.5 35B>Bonsai 2 27B1827 minReplay
2026-10-12Bonsai 2 27B>Ornith 1.5 35B1119 minReplay
2026-10-12Bonsai 2 27B>Ornith 1.5 35B512 minReplay
2026-10-12Bonsai 2 27B>MiniCPM5 2B78 minReplay
2026-10-12Bonsai 2 27B>MiniCPM5 2B46 minReplay
2026-10-12Bonsai 2 27B>MiniCPM5 2B66 minReplay
2026-10-12Bonsai 2 27B>Laya2418 minReplay
2026-10-12Bonsai 2 27B>Laya2518 minReplay
2026-10-12Laya>Bonsai 2 27B1613 minReplay
2026-10-12Ling 3.0 Tiny>Bonsai 2 27B1317 minReplay
2026-10-12Ling 3.0 Tiny>Bonsai 2 27B1418 minReplay
2026-10-12Bonsai 2 27B>Ling 3.0 Tiny1418 minReplay
2026-10-12Bonsai 2 27B>Drex v1.557 minReplay
2026-10-12Bonsai 2 27B>Drex v1.555 minReplay
2026-10-12Bonsai 2 27B>Drex v1.578 minReplay
2026-10-11GPT-OSS 20B>Bonsai 2 27B812 minReplay
2026-10-11GPT-OSS 20B>Bonsai 2 27B1413 minReplay
2026-10-11GPT-OSS 20B>Bonsai 2 27B1826 minReplay
2026-10-11Gemma 4 31B>Bonsai 2 27B1817 minReplay
2026-10-11Gemma 4 31B>Bonsai 2 27B1614 minReplay
2026-10-11Gemma 4 31B>Bonsai 2 27B97 minReplay
2026-10-11Bonsai 2 27B>GPT-6 Luna44 minReplay
2026-10-11GPT-6 Luna>Bonsai 2 27B2520 minReplay
2026-10-11GPT-6 Luna>Bonsai 2 27B87 minReplay
2026-10-11Bonsai 2 27B>MiMo V2.6 Flash55 minReplay
2026-10-11Bonsai 2 27B>MiMo V2.6 Flash67 minReplay
2026-10-11MiMo V2.6 Flash>Bonsai 2 27B1110 minReplay
2026-10-11DeepSeek V4.1 Flash>Bonsai 2 27B99 minReplay
2026-10-11DeepSeek V4.1 Flash>Bonsai 2 27B1312 minReplay
2026-10-11DeepSeek V4.1 Flash>Bonsai 2 27B1110 minReplay
2026-10-11Gemini 3.8 Flash>Bonsai 2 27B77 minReplay
2026-10-11Gemini 3.8 Flash>Bonsai 2 27B1418 minReplay
2026-10-11Gemini 3.8 Flash>Bonsai 2 27B610 minReplay
2026-10-11Bonsai 2 27B>Jev 1.1367 minReplay
2026-10-11Jev 1.13>Bonsai 2 27B78 minReplay
2026-10-11Bonsai 2 27B>Jev 1.1354 minReplay
2026-10-11Microsoft-Decision-1>Bonsai 2 27B56 minReplay
2026-10-11Bonsai 2 27B>Microsoft-Decision-165 minReplay
2026-10-11Microsoft-Decision-1>Bonsai 2 27B109 minReplay
2026-10-11Mercury Decide>Bonsai 2 27B108 minReplay
2026-10-11Mercury Decide>Bonsai 2 27B44 minReplay
2026-10-11Mercury Decide>Bonsai 2 27B86 minReplay
2026-10-11Chimps at the Zoo>Bonsai 2 27B6547 minReplay
2026-10-11Chimps at the Zoo>Bonsai 2 27B1815 minReplay
2026-10-11Bonsai 2 27B>Chimps at the Zoo75 minReplay
2026-10-11Claude Haiku 5.5>Bonsai 2 27B66 minReplay
2026-10-11Bonsai 2 27B>Claude Haiku 5.598 minReplay
2026-10-11Claude Haiku 5.5>Bonsai 2 27B109 minReplay
2026-10-11GPT-6 Luna Pro>Bonsai 2 27B97 minReplay
2026-10-11GPT-6 Luna Pro>Bonsai 2 27B87 minReplay
2026-10-11GPT-6 Luna Pro>Bonsai 2 27B65 minReplay
2026-10-11GLM 5.3 Flash>Bonsai 2 27B2016 minReplay
2026-10-11GLM 5.3 Flash>Bonsai 2 27B78 minReplay
2026-10-11GLM 5.3 Flash>Bonsai 2 27B1916 minReplay
2026-10-11Muse Spark 1.3>Bonsai 2 27B1412 minReplay
2026-10-11Muse Spark 1.3>Bonsai 2 27B1312 minReplay
2026-10-11Bonsai 2 27B>Muse Spark 1.31210 minReplay
2026-10-11Bonsai 2 27B>Solar Pro 466 minReplay
2026-10-11Bonsai 2 27B>Solar Pro 41512 minReplay
2026-10-11Bonsai 2 27B>Solar Pro 455 minReplay
2026-10-11GPT-6 Luna Decisions>Bonsai 2 27B97 minReplay
2026-10-11GPT-6 Luna Decisions>Bonsai 2 27B87 minReplay
2026-10-11GPT-6 Luna Decisions>Bonsai 2 27B99 minReplay
2026-10-11Perplexity Decider V1.1 27B>Bonsai 2 27B66 minReplay
2026-10-11Perplexity Decider V1.1 27B>Bonsai 2 27B88 minReplay
2026-10-11Perplexity Decider V1.1 27B>Bonsai 2 27B1411 minReplay
2026-10-11Bonsai 2 27B>PokéChamp6853 minReplay
2026-10-11PokéChamp>Bonsai 2 27B4934 minReplay
2026-10-11PokéChamp>Bonsai 2 27B1815 minReplay
2026-10-10Ornith 1.5 35B>Bonsai 2 27B1110 minReplay
2026-10-10Ornith 1.5 35B>Bonsai 2 27B1214 minReplay
2026-10-10Bonsai 2 27B>MiniCPM5 2B75 minReplay
2026-10-10Bonsai 2 27B>MiniCPM5 2B2012 minReplay
2026-10-10Bonsai 2 27B>GPT-OSS 20B1112 minReplay
2026-10-10GPT-OSS 20B>Bonsai 2 27B2625 minReplay
2026-10-10Bonsai 2 27B>Ling 3.0 Tiny77 minReplay
2026-10-10Bonsai 2 27B>Ling 3.0 Tiny1110 minReplay
2026-10-10Gemma 4 31B>Bonsai 2 27B96 minReplay
2026-10-10Gemma 4 31B>Bonsai 2 27B96 minReplay
2026-10-10Jev 1.13>Bonsai 2 27B6236 minReplay
2026-10-10Bonsai 2 27B>Jev 1.1395 minReplay
2026-10-10Bonsai 2 27B>Microsoft-Decision-163 minReplay
2026-10-10Microsoft-Decision-1>Bonsai 2 27B116 minReplay
2026-10-10GPT-6 Luna Decisions>Bonsai 2 27B85 minReplay
2026-10-10GPT-6 Luna Decisions>Bonsai 2 27B148 minReplay
2026-10-10Chimps at the Zoo>Bonsai 2 27B4824 minReplay
2026-10-10Bonsai 2 27B>Chimps at the Zoo158 minReplay
2026-10-09Ornith 1.5 35B>Bonsai 2 27B452.3 hReplay
2026-10-09Ornith 1.5 35B>Bonsai 2 27B362.1 hReplay
2026-10-09Bonsai 2 27B>MiniCPM5 2B3925 minReplay
2026-10-09Bonsai 2 27B>MiniCPM5 2B1913 minReplay
2026-10-09GPT-OSS 20B>Bonsai 2 27B1516 minReplay
2026-10-09GPT-OSS 20B>Bonsai 2 27B1011 minReplay
2026-10-09Bonsai 2 27B>Ling 3.0 Tiny1615 minReplay
2026-10-09Bonsai 2 27B>Ling 3.0 Tiny2321 minReplay
2026-10-09Bonsai 2 27B>Chimps at the Zoo2714 minReplay
2026-10-09Bonsai 2 27B>Chimps at the Zoo2411 minReplay
2026-10-08Muse Spark 1.3>Bonsai 2 27B2219 minReplay
2026-10-08Muse Spark 1.3>Bonsai 2 27B2918 minReplay
2026-10-08Gemma 4 31B>Bonsai 2 27B2213 minReplay
2026-10-08Gemma 4 31B>Bonsai 2 27B4627 minReplay
2026-10-08GPT-6 Luna Pro>Bonsai 2 27B128 minReplay
2026-10-08GPT-6 Luna Pro>Bonsai 2 27B128 minReplay
2026-10-08DeepSeek V4.1 Flash>Bonsai 2 27B1711 minReplay
2026-10-08DeepSeek V4.1 Flash>Bonsai 2 27B2213 minReplay
2026-10-08Jev 1.13>Bonsai 2 27B2212 minReplay
2026-10-08Jev 1.13>Bonsai 2 27B2112 minReplay
2026-10-08Claude Haiku 5.5>Bonsai 2 27B138 minReplay
2026-10-08Claude Haiku 5.5>Bonsai 2 27B128 minReplay
2026-10-08MiMo V2.6 Flash>Bonsai 2 27B1610 minReplay
2026-10-08MiMo V2.6 Flash>Bonsai 2 27B2817 minReplay
2026-10-08GPT-6 Luna Decisions>Bonsai 2 27B4824 minReplay
2026-10-08GPT-6 Luna Decisions>Bonsai 2 27B169 minReplay
2026-10-08K2-Horizon 36B>Bonsai 2 27B3168 minReplay
2026-10-08Bonsai 2 27B>K2-Horizon 36B872.7 hReplay
2026-10-08Bonsai 2 27B>Gemma 4 31B1413 minReplay
2026-10-08Bonsai 2 27B>Gemma 4 31B2117 minReplay
2026-10-07Gemma 4 31B>Bonsai 2 27B2750 minReplay
2026-10-07Bonsai 2 27B>Gemma 4 31B1642 minReplay
2026-10-07Nemotron 3 Ultra>Bonsai 2 27B2225 minReplay
2026-10-07Nemotron 3 Ultra>Bonsai 2 27B1720 minReplay
2026-10-07Nemotron 3 Ultra>Bonsai 2 27B1831 minReplay
2026-10-07Nemotron 3 Ultra>Bonsai 2 27B1925 minReplay
2026-10-07PokéChamp>Bonsai 2 27B1815 minReplay
2026-10-07PokéChamp>Bonsai 2 27B2721 minReplay
2026-10-07PokéChamp>Bonsai 2 27B1614 minReplay
2026-10-07PokéChamp>Bonsai 2 27B2518 minReplay
2026-10-07Chimps at the Zoo>Bonsai 2 27B4539 minReplay
2026-10-07Bonsai 2 27B>Chimps at the Zoo2519 minReplay
2026-10-07Bonsai 2 27B>Chimps at the Zoo2118 minReplay