TIS-100 Bench: Signal Edge Detector
Compare each value with its predecessor and flag changes of at least 10. The first output is always zero.
In TIS-100 Bench, 16 LLMs each had one attempt at this TIS-100 puzzle, with 10 minutes to find a working program and keep improving it, without seeing the community leaderboard. 14 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 | 102 | 204 | GPT-6.1 Sol · +100% |
| Nodes | 4 | 4 | DeepSeek V4.1 Flash Ties record |
| Instructions | 14 | 15 | Fable 5.1 · +7% |
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 |
|---|---|---|---|---|---|
| 01GPT-6.1 Sol | Yes | 204 | 4 | 18 | — |
| 02Opus 5.5 | Yes | 204 | 4 | 15 | — |
| 03GPT-6 Astra | Yes | 214 | 4 | 15 | — |
| 04Sonnet 5.5 | Yes | 214 | 4 | 15 | — |
| 05Fable 5.1 | Yes | 222 | 4 | 15 | — |
| 06Gemini 3.8 Flash | Yes | 231 | 4 | 15 | — |
| 07Gemini 4 Argon | Yes | 241 | 4 | 15 | — |
| 08Haiku 5.5 | Yes | 241 | 4 | 18 | — |
| 09Kimi K3.1 Pro | Yes | 244 | 4 | 22 | — |
| 10Muse Spark 1.3 | Yes | 244 | 4 | 16 | — |
| 11DeepSeek V4.1 Flash | Yes | 245 | 4 | 20 | — |
| 12Muse Wonder 1.0 | Yes | 245 | 4 | 15 | — |
| 13GPT-6 Luna | Yes | 280 | 4 | 20 | — |
| 14Grok 4.7 | Yes | 322 | 5 | 21 | — |
| 15GLM-5.3 | No | — | — | — | — |
| 16MiMo-V2.6-Pro | 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.
GPT-6.1 Sol
204 cycles · 6 nodes · 30 instructions32050.204-6-30.txt on GitHub

MOV UP ACC
MOV ACC RIGHT
MOV ACC DOWNMOV LEFT ACC
SUB LEFT
MOV ACC DOWNMOV UP NIL
MOV 0 DOWN
L: MOV UP ACC
SUB UP
MOV ACC DOWN
JMP LL: MOV UP ACC
JGZ P
NEG
P: SUB 10
JLZ Z
MOV 1 DOWN
JMP L
Z: MOV 0 DOWNL: MOV UP ACC
JGZ P
NEG
P: SUB 10
JLZ Z
MOV 1 RIGHT
JMP L
Z: MOV 0 RIGHTMOV LEFT DOWN
MOV UP DOWNGPT-6 Astra
238 cycles · 4 nodes · 18 instructions32050.238-4-18.txt on GitHub

MOV UP ACC
L:MOV ACC DOWN
MOV UP ACC
MOV ACC DOWN
JMP LMOV UP ACC
SUB UP
MOV ACC RIGHTMOV LEFT ACC
JGZ P
NEG
P:MOV ACC DOWNZ:MOV 0 DOWN
L:MOV UP ACC
SUB 10
JLZ Z
MOV 1 DOWN
JMP LGemini 4 Argon
272 cycles · 4 nodes · 15 instructions32050.272-4-15.txt on GitHub

MOV UP, DOWNMOV UP, ACC
MOV ACC, DOWN
MOV ACC, DOWNSUB UP
MOV ACC, RIGHT
MOV UP, ACCL: MOV LEFT, ACC
JGZ P
NEG
P: SUB 10
JLZ Z
MOV 1, DOWN
JMP L
Z: MOV 0, DOWN