TIS-100 Bench: Differential Converter

20176 Easy Routing & synchronization Arithmetic construction Built-in puzzle · TIS-100 Bench report →

Read two streams and emit both differences, A−B and B−A. Both outputs must stay synchronized with the same input pair.

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. 16 of 16 models solved it. No model found a new Pareto-frontier point beyond the community leaderboard.

16 / 16models solved it
143best LLM cycles (126 community)
0new frontier points

Records

ScoreCommunity bestBest LLMModel
Cycles126143Fable 5.1 · +13%
Nodes55DeepSeek V4.1 Flash Ties record
Instructions1010DeepSeek V4.1 Flash Ties record

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 ↙
Loading interactive Pareto explorer…
Community leaderboardModel frontier pointNew frontier point
Select a point with the mouse or keyboard to inspect its scores and program.

Search over time

Loading puzzle trajectories…

Every model

ModelSolvedBest cyclesBest nodesBest instructionsNew frontier points
01Fable 5.1Yes143510—
02GPT-6 AstraYes163510—
03Opus 5.5Yes164510—
04Gemini 4 ArgonYes165510—
05GPT-6.1 SolYes182510—
06Gemini 3.8 FlashYes200510—
07DeepSeek V4.1 FlashYes201510—
08GLM-5.3Yes201510—
09Haiku 5.5Yes201510—
10Muse Spark 1.3Yes201510—
11Muse Wonder 1.0Yes201510—
12Sonnet 5.5Yes201510—
13Grok 4.7Yes203510—
14MiMo-V2.6-ProYes203510—
15Kimi K3.1 ProYes238510—
16GPT-6 LunaYes238611—

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.

Fable 5.1 Fewest cycles

143 cycles · 9 nodes · 25 instructions20176.143-9-25.txt on GitHub

Fable 5.1 program running on 20176 in TIS-100
Node 0
MOV RIGHT, DOWN
Node 1
MOV UP, DOWN
MOV UP, ACC
SUB RIGHT
MOV ACC, LEFT
Node 2
MOV UP, DOWN
MOV UP, LEFT
Node 4
MOV UP, DOWN
Node 5
MOV UP, ACC
SUB RIGHT
MOV ACC, RIGHT
MOV ACC, DOWN
Node 6
MOV UP, LEFT
MOV LEFT, ACC
NEG
MOV ACC, DOWN
Node 7
MOV UP, RIGHT
Node 8
MOV UP, DOWN
MOV LEFT, ACC
MOV ACC, RIGHT
MOV ACC, DOWN
Node 9
MOV UP, DOWN
MOV LEFT, ACC
NEG
MOV ACC, DOWN

Gemini 3.8 Flash Fewest nodes

200 cycles · 5 nodes · 11 instructions20176.200-5-11.txt on GitHub

Gemini 3.8 Flash program running on 20176 in TIS-100
Node 1
MOV UP, ACC
SUB RIGHT
MOV ACC, DOWN
Node 2
MOV UP, LEFT
Node 5
MOV UP, DOWN
Node 8
MOV UP, ACC
MOV ACC, RIGHT
MOV ACC, DOWN
Node 9
SUB LEFT
MOV ACC, DOWN
MOV 0, ACC

DeepSeek V4.1 Flash Fewest instructions

240 cycles · 5 nodes · 10 instructions20176.240-5-10.txt on GitHub

DeepSeek V4.1 Flash program running on 20176 in TIS-100
Node 1
MOV UP, RIGHT
Node 2
MOV UP, ACC
SUB LEFT
MOV ACC, DOWN
Node 6
MOV UP, DOWN
Node 8
MOV RIGHT, DOWN
Node 9
MOV UP, ACC
MOV ACC, DOWN
NEG
MOV ACC, LEFT