TIS-100 Bench: Sequence Sorter
Sort each zero-terminated sequence before emitting it. The grid has to hold and reorder values with very little local state.
In TIS-100 Bench, 16 LLMs each had one attempt at this TIS-100 puzzle, with 30 minutes to find a working program and keep improving it, without seeing the community leaderboard. 8 of 16 models solved it. No model found a new Pareto-frontier point beyond the community leaderboard.
Records
| Score | Community best | Best LLM | Model |
|---|---|---|---|
| Cycles | 463 | 1,475 | Kimi K3.1 Pro · +219% |
| Nodes | 3 | 4 | Fable 5.1 · +33% |
| Instructions | 24 | 39 | GPT-6.1 Sol · +62% |
Community records from the community leaderboard when we froze it. A record counts only after fresh-game replay and 100,000 passing random tests.
Pareto frontier
Pareto explorer
Lower is better ↙Search over time
Loading puzzle trajectories…
The search over time
Best in-game score so far, before independent testing. Each score can come from a different program, and lines stop when the episode ends.
Every model
| Model | Solved | Best cycles | Best nodes | Best instructions | New frontier points |
|---|---|---|---|---|---|
| 01Kimi K3.1 Pro | Yes | 1,475 | 4 | 41 | — |
| 02GPT-6 Astra | Yes | 1,511 | 4 | 45 | — |
| 03GPT-6.1 Sol | Yes | 1,520 | 4 | 39 | — |
| 04Gemini 4 Argon | Yes | 1,690 | 4 | 39 | — |
| 05Sonnet 5.5 | Yes | 1,821 | 5 | 51 | — |
| 06Opus 5.5 | Yes | 1,872 | 4 | 39 | — |
| 07Gemini 3.8 Flash | Yes | 2,132 | 5 | 55 | — |
| 08Fable 5.1 | Yes | 2,982 | 4 | 48 | — |
| 09DeepSeek V4.1 Flash | No | — | — | — | — |
| 10GLM-5.3 | No | — | — | — | — |
| 11Grok 4.7 | No | — | — | — | — |
| 12Haiku 5.5 | No | — | — | — | — |
| 13MiMo-V2.6-Pro | No | — | — | — | — |
| 14Muse Spark 1.3 | No | — | — | — | — |
| 15Muse Wonder 1.0 | No | — | — | — | — |
| 16GPT-6 Luna | No | — | — | — | — |
Best scores can come from different programs. New frontier points are not matched or beaten on all three scores by any community program.
Best model programs
These programs are in TIS-100 save format and released under the MIT License. All of this puzzle’s model programs are on GitHub and Hugging Face.
Kimi K3.1 Pro
1,475 cycles · 4 nodes · 42 instructions63534.1475-4-42.txt on GitHub

MOV UP,DOWNI: MOV UP,ACC
SAV
MOV ACC,RIGHT
JEZ I
JMP S
M: MOV DOWN,ACC
MOV ACC,RIGHT
S: MOV DOWN,ACC
MOV ACC,DOWN
MOV ACC,RIGHT
JRO RIGHT
SWP
MOV ACC,DOWNMOV 0,UP
I: MOV LEFT,ACC
JGZ C
MOV -1,UP
JMP R
M: MOV -5,LEFT
SWP
MOV LEFT,UP
C: SAV
SUB LEFT
JGZ M
MOV 1,LEFT
R: MOV UP,ACC
MOV ACC,DOWN
JNZ RI: MOV 999,LEFT
JMP R
P: MOV ACC,LEFT
R: MOV UP,ACC
JGZ P
JEZ R
O: MOV LEFT,ACC
SUB 999
JEZ E
ADD 999
MOV ACC,DOWN
JMP O
E: MOV 0,DOWNGemini 4 Argon
1,729 cycles · 4 nodes · 39 instructions63534.1729-4-39.txt on GitHub

L: MOV UP, ACC
MOV ACC, DOWN
JNZ L
MOV DOWN, NILP: MOV 0, RIGHT
MOV ANY, ACC
MOV ACC, RIGHT
JNZ L
MOV 0, UP
JMP P
L: MOV ANY, ACC
MOV ACC, RIGHT
JNZ L
MOV 0, DOWN MOV LEFT, ACC
L1:MOV ACC, UP
MOV LEFT, ACC
JNZ L1
MOV UP, ACC
MOV ACC, DOWN
JEZ E
L2:MOV UP, ACC
E: MOV ACC, DOWN
MOV ACC, DOWN
JNZ L2 MOV UP, ACC
L: SAV
SUB UP
JLZ LS
MOV UP, ACC
JEZ O
SWP
MOV ACC, LEFT
JMP S
LS:MOV UP, LEFT
S: SWP
JMP L
O: SWP
MOV ACC, DOWN