SpaceChem Bench: Slightly Different

Inputs: β Hydrogen (H2). Output: ω Hydrogen (H2) ×10.

In SpaceChem Bench, 16 LLMs each had one attempt at this SpaceChem puzzle, with 10 minutes to build a working solution and keep optimizing it, without seeing the community archive. 15 of 16 models solved it. Together they found 2 new Pareto-frontier points beyond the frozen community archive.

15 / 16models solved it
47best LLM cycles (45 community)
2new frontier points

Records

ScoreCommunity bestBest LLMModel
Cycles4547GPT-6.1 Sol · +4%
Reactors11DeepSeek V4.1 Flash Ties record
Symbols66DeepSeek V4.1 Flash Ties record

Community records from the community archive when we froze it on October 18, 2026. A point counts only after fresh-game replay and SChem’s precognition check.

Pareto frontier

Pareto explorer

Lower is better ↙
Loading interactive Pareto explorer…
Community archiveModel frontier pointNew frontier point
Select a point with the mouse or keyboard to inspect its scores and reactor blueprint.

Search over time

Loading puzzle trajectories…

Every model

ModelSolvedBest cyclesBest reactorsBest symbolsNew frontier points
01GPT-6 AstraYes48162
02GPT-6.1 SolYes47161
03Muse Wonder 1.0Yes52161
04Fable 5.1Yes5216—
05Gemini 4 ArgonYes5216—
06Opus 5.5Yes5216—
07DeepSeek V4.1 FlashYes5716—
08Sonnet 5.5Yes5716—
09Kimi K3.1 ProYes5916—
10Gemini 3.8 FlashYes6216—
11Grok 4.7Yes6216—
12Muse Spark 1.3Yes6316—
13GPT-6 LunaYes10716—
14Haiku 5.5Yes11616—
15GLM-5.3Yes11716—
16MiMo-V2.6-ProNo————

Best scores can come from different solutions. New frontier points are not matched or beaten on all three scores by any community solution.

Best model solutions

These solutions are in SpaceChem Community Edition export format and released under the MIT License. All of this puzzle’s model solutions are on GitHub and Hugging Face.

GPT-6.1 Sol Fewest cyclesFewest reactors

47 cycles · 1 reactor · 12 symbolsresearch_tutorial_1.47-1-12.txt on GitHub

Layout

α β ψ ω 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 → ← → ← S→ S↓ G/D G/D CW OUT ω IN β CW G/D G/D

Running in the game

GPT-6.1 Sol solution running on research_tutorial_1 in SpaceChem
SpaceChem Community Edition export (research_tutorial_1.47-1-12.txt)
SOLUTION:Slightly Different,Player,0-0-0
COMPONENT:'empty-research-reactor',2,0,''
MEMBER:'instr-start',0,0,128,4,5,0,0
MEMBER:'instr-start',90,0,32,2,4,0,0
MEMBER:'instr-arrow',0,0,64,2,5,0,0
MEMBER:'instr-arrow',180,0,64,6,5,0,0
MEMBER:'instr-grab',-1,0,128,2,5,0,0
MEMBER:'instr-rotate',-1,0,128,3,5,0,0
MEMBER:'instr-input',-1,1,128,5,5,0,0
MEMBER:'instr-grab',-1,0,128,6,5,0,0
MEMBER:'instr-arrow',0,0,16,2,5,0,0
MEMBER:'instr-arrow',180,0,16,6,5,0,0
MEMBER:'instr-grab',-1,0,32,2,5,0,0
MEMBER:'instr-output',-1,1,32,3,5,0,0
MEMBER:'instr-rotate',-1,0,32,5,5,0,0
MEMBER:'instr-grab',-1,0,32,6,5,0,0

GPT-6 Astra Fewest symbols

115 cycles · 1 reactor · 6 symbolsresearch_tutorial_1.115-1-6.txt on GitHub

Layout

α β ψ ω 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 → ← S↓ S← IN β G/D G/D OUT ω

Running in the game

GPT-6 Astra solution running on research_tutorial_1 in SpaceChem
SpaceChem Community Edition export (research_tutorial_1.115-1-6.txt)
SOLUTION:Slightly Different,Player,0-0-0
COMPONENT:'empty-research-reactor',2,0,''
MEMBER:'instr-start',90,0,128,0,4,0,0
MEMBER:'instr-start',180,0,32,1,0,0,0
MEMBER:'instr-arrow',0,0,64,0,5,0,0
MEMBER:'instr-arrow',180,0,64,6,5,0,0
MEMBER:'instr-input',-1,1,128,0,5,0,0
MEMBER:'instr-grab',-1,0,128,1,5,0,0
MEMBER:'instr-grab',-1,0,128,6,5,0,0
MEMBER:'instr-output',-1,1,32,0,0,0,0