TIS-100 Bench: Back-Reference Reifier

NEXUS.15.897.9 Medium Memory & interpreters Streaming state Built-in puzzle · TIS-100 Bench report →

Expand references to one of the last four input values, or consume a new value on command zero. History must survive repeated lookups.

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

13 / 16models solved it
330best LLM cycles (219 community)
0new frontier points

Records

ScoreCommunity bestBest LLMModel
Cycles219330Opus 5.5 · +51%
Nodes44GPT-6 Astra Ties record
Instructions2131Sonnet 5.5 · +48%

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
01Opus 5.5Yes330439—
02GPT-6 AstraYes350434—
03Gemini 4 ArgonYes416444—
04GPT-6.1 SolYes434442—
05Haiku 5.5Yes570549—
06Kimi K3.1 ProYes577633—
07Fable 5.1Yes591535—
08Gemini 3.8 FlashYes624665—
09Muse Wonder 1.0Yes640665—
10Sonnet 5.5Yes681431—
11Grok 4.7Yes744650—
12DeepSeek V4.1 FlashYes1,012557—
13Muse Spark 1.3Yes1,274869—
14GLM-5.3No————
15MiMo-V2.6-ProNo————
16GPT-6 LunaNo————

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.

Opus 5.5 Fewest cycles

330 cycles · 6 nodes · 55 instructionsNEXUS.15.897.9.330-6-55.txt on GitHub

Opus 5.5 program running on NEXUS.15.897.9 in TIS-100
Node 1
MOV UP, ACC
NEG
ADD ACC
ADD 1
MOV ACC, DOWN
Node 2
MOV UP, ACC
MOV ACC, DOWN
MOV ACC, DOWN
Node 5
S: MOV UP, ACC
MOV ACC, DOWN
JRO ACC
MOV 10, RIGHT
JMP S
MOV 7, RIGHT
JMP S
MOV 1, RIGHT
JMP S
MOV 8, RIGHT
JMP S
MOV 8, RIGHT
Node 6
J: JRO LEFT
SWP
MOV ACC, DOWN
MOV ACC, DOWN
MOV UP, NIL
MOV UP, ACC
JMP J
MOV ACC, DOWN
MOV UP, NIL
JMP S
MOV UP, DOWN
S: SWP
MOV ACC, DOWN
MOV UP, ACC
Node 8
S: MOV UP, ACC
JRO ACC
MOV 7, RIGHT
JMP S
MOV 7, RIGHT
JMP S
MOV 7, RIGHT
JMP S
MOV 5, RIGHT
JMP S
MOV 1, RIGHT
Node 9
J: JRO LEFT
SWP
MOV ACC, DOWN
MOV UP, ACC
JMP J
MOV ACC, DOWN
JMP S
MOV UP, DOWN
S: SAV
MOV UP, ACC

Opus 5.5 Fewest nodes

470 cycles · 4 nodes · 43 instructionsNEXUS.15.897.9.470-4-43.txt on GitHub

Opus 5.5 program running on NEXUS.15.897.9 in TIS-100
Node 1
MOV UP, ACC
ADD ACC
ADD ACC
ADD 17
MOV ACC, RIGHT
Node 2
MOV LEFT, ACC
MOV ACC, DOWN
SUB 10
JGZ K
SUB 1
K: NEG
ADD 8
MOV ACC, DOWN
MOV UP, DOWN
Node 6
F: MOV UP, DOWN
S: SWP
L: MOV ACC, DOWN
JRO UP
MOV UP, ACC
MOV ACC, DOWN
MOV UP, DOWN
JMP S
MOV UP, ACC
SWP
MOV ACC, DOWN
MOV UP, DOWN
JMP L
MOV ACC, DOWN
MOV UP, ACC
Node 9
J: JRO UP
SWP
MOV ACC, DOWN
MOV UP, ACC
JMP J
MOV ACC, DOWN
SAV
MOV UP, ACC
JMP J
P: SAV
MOV UP, ACC
MOV UP, DOWN
JMP J
JMP P

Sonnet 5.5 Fewest instructions

1,017 cycles · 5 nodes · 31 instructionsNEXUS.15.897.9.1017-5-31.txt on GitHub

Sonnet 5.5 program running on NEXUS.15.897.9 in TIS-100
Node 1
MOV UP, ACC
JNZ A
SUB 5
A: MOV ACC, RIGHT
Node 2
MOV LEFT, DOWN
MOV ACC, DOWN
SWP
MOV ACC, DOWN
MOV RIGHT, DOWN
MOV RIGHT, DOWN
MOV ACC, RIGHT
MOV UP, ACC
MOV ACC, DOWN
Node 3
MOV ACC, LEFT
SWP
MOV ACC, LEFT
MOV LEFT, ACC
Node 6
S: MOV UP, ACC
SAV
L1: ADD 1
JEZ PK
MOV UP, NIL
JMP L1
PK: MOV UP, DOWN
SWP
ADD 5
L2: JEZ S
MOV UP, NIL
SUB 1
JMP L2
Node 9
MOV UP, DOWN