TIS-100 Bench: Puzzles

All 55 TIS-100 puzzles from TIS-100 Bench: LLMs Push the Pareto Frontier in Programming Puzzle Games, with each puzzle’s difficulty, mechanisms, specification summary, community leaderboard cycle record, and the fastest of the 16 evaluated LLMs. Select any puzzle for its full records across cycles, nodes, and instructions, Pareto chart, and best model-written assembly programs on GitHub and Hugging Face.

133new frontier points
18 / 55puzzles with new frontier points
3outright cycle records

Built-in puzzles 47

PuzzleTierMechanismsCommunitycyclesBest LLMcyclesSolvedmodelsNewpoints
Copy two input streams to their matching outputs. The first challenge is simply routing values through a damaged processor grid.Easy
Routing & synchronization
8383Gemini 4 Argon+1516/16—
Double each input value. A tiny arithmetic task exposes the cost of getting values to and from the working node.Easy
Arithmetic construction Routing & synchronization
8484Gemini 4 Argon+1316/16—
Read two streams and emit both differences, A−B and B−A. Both outputs must stay synchronized with the same input pair.Easy
Routing & synchronization Arithmetic construction
126143Fable 5.116/16—
Classify each input as positive, zero, or negative. Every input produces three outputs: one 1 and two 0s.Easy
Routing & synchronization
188227GPT-6 Astra15/16—
A selector chooses input A, input B, or their sum. All three input streams must be consumed in step.Easy
Routing & synchronization Arithmetic construction
145202Muse Spark 1.315/16—
Turn each input pair into a sequence: the smaller value, the larger value, then a zero terminator.Easy
Sequence algorithms
8687Muse Wonder 1.014/16—
Compute the sum and length of each zero-terminated sequence. The delimiter resets both running totals.Easy
Sequence algorithms Arithmetic construction
112195Muse Wonder 1.014/16—
Compare each value with its predecessor and flag changes of at least 10. The first output is always zero.Easy
Streaming state Arithmetic construction
102204Opus 5.5+114/16—
Monitor four streams and report which one changes from 0 to 1. Previous values must be remembered independently.Easy
Streaming state Routing & synchronization
122169GPT-6.1 Sol+214/16—
Detect three consecutive zeros in a stream. Overlapping matches make the update and reset rules matter.Easy
Streaming state
82126GPT-6.1 Sol15/16—
Find the minimum and maximum of each zero-terminated sequence, maintaining two summaries until its end.Medium
Sequence algorithms Streaming state
185227GPT-6 Astra15/16—
Reverse zero-terminated sequences using the available stack memory. Reading, storing, and draining must agree on sequence boundaries.Easy
Sequence algorithms Memory & interpreters
184240Gemini 4 Argon15/16—
Multiply two inputs without a multiply instruction. Repeated addition, loop unrolling, and parallel work offer different tradeoffs.Easy
Arithmetic construction
198436Gemini 4 Argon15/16—
Draw a fixed target image by emitting coordinates and pixel values. Compact generation competes with the number of output operations.Easy
Image generation
1,1861,187Gemini 4 Argon+214/16—
Draw a second fixed target image. The task is to encode its spatial pattern within the small programs available on the grid.Easy
Image generation
1,1501,151Gemini 4 Argon+214/16—
Read a rectangle's position, width, and height, then draw it. Coordinate loops must stop at exactly the requested boundaries.Medium
Image generation Arithmetic construction
511556Gemini 4 Argon+113/16—
Turn successive input values into adjacent vertical bars. Each value controls a bar's length and advances the drawing position.Easy
Image generation Streaming state
1,2012,080GPT-6.1 Sol14/16—
Maintain sums of the last three and last five inputs, starting with zero-filled history. Old values must leave each window at the right time.Easy
Streaming state Arithmetic construction
200317Fable 5.112/16—
Divide A by B and output both quotient and remainder, using an instruction set without division.Medium
Arithmetic construction
8621,376GPT-6 Astra16/162
Store a zero-terminated sequence, then answer index lookups. Values must remain available for later requests.Medium
Memory & interpreters Sequence algorithms
291989Sonnet 5.515/16—
Sort each zero-terminated sequence before emitting it. The grid has to hold and reorder values with very little local state.Medium
Sequence algorithms Memory & interpreters
4631,475Kimi K3.1 Pro8/16—
Decode length–color pairs into an image, wrapping a run onto the next row when it crosses the image edge.Medium
Image generation Sequence algorithms
1,1781,314Gemini 4 Argon15/16—
Sequence Merger NEXUS.00.526.6
Merge two already-sorted, zero-terminated sequences. Advancing one input must preserve the unconsumed value from the other.
Medium
Sequence algorithms Routing & synchronization
279347GPT-6 Astra12/16—
Sum every integer from 1 through the input value. Construct the arithmetic using the available primitive operations.Easy
Arithmetic construction
264907GPT-6 Astra15/16—
Sequence Range Limiter NEXUS.02.981.2
Read lower and upper bounds, then clamp every element of a zero-terminated sequence to that range.
Medium
Sequence algorithms Streaming state
149283Haiku 5.514/16—
Signal Error Corrector NEXUS.03.176.9
Copy two streams, replacing a negative value with its partner from the other stream. At most one partner is negative at a time.
Easy
Routing & synchronization
138201Gemini 4 Argon16/16—
Subsequence Extractor NEXUS.04.340.5
Extract an inclusive range from a zero-terminated sequence. Values outside the requested range still have to be consumed.
Medium
Sequence algorithms Streaming state
94102Gemini 4 Argon15/16—
Signal Prescaler NEXUS.05.647.1
Emit each input divided by 2, 4, and 8. Sharing intermediate work can save instructions and communication.
Medium
Arithmetic construction Routing & synchronization
304429Opus 5.515/16—
Signal Averager NEXUS.06.786.0
Average two inputs and round down. Neither averaging nor division is a native instruction.
Easy
Arithmetic construction Routing & synchronization
4321,181Gemini 4 Argon12/16—
Submaximum Selector NEXUS.07.050.0
Select the second-largest of four inputs. Keeping the top two candidates is enough; fully sorting is another option.
Hard
Sequence algorithms Routing & synchronization
336414Kimi K3.1 Pro16/164
Decimal Decomposer NEXUS.08.633.9
Split an input into hundreds, tens, and ones digits, sending each digit to its own output.
Medium
Arithmetic construction Routing & synchronization
394579Grok 4.715/165
Find the most frequent value in each zero-terminated sequence. Output zero when the maximum frequency is tied.Hard
Sequence algorithms Memory & interpreters
287472Gemini 4 Argon4/162
Sequence Normalizer NEXUS.10.656.5
Find a sequence's minimum, then subtract it from every element. The terminator is −1, so zero is valid data.
Hard
Sequence algorithms Memory & interpreters
317494Opus 5.515/16—
Image Test Pattern 3 NEXUS.11.711.2
Generate a third fixed target image. Reuse spatial regularities to fit the drawing procedure into the node limits.
Medium
Image generation
8821,520GPT-6 Astra12/16—
Image Test Pattern 4 NEXUS.12.534.4
Generate a fourth fixed target image. Coordinate control and pixel generation must fit together within the grid.
Medium
Image generation
1,1661,222Opus 5.515/16—
Spatial Path Viewer NEXUS.13.370.9
Draw connected line segments from angle–length pairs, starting at the upper-left corner. Each segment updates the next starting point.
Medium
Image generation Streaming state
744975Opus 5.513/161
Character Terminal NEXUS.14.781.3
Render character codes into an image, treating zero as a newline. Character patterns and cursor position both need representation.
Hard
Image generation Memory & interpreters
578654Fable 5.111/1611
Back-Reference Reifier NEXUS.15.897.9
Expand references to one of the last four input values, or consume a new value on command zero. History must survive repeated lookups.
Medium
Memory & interpreters Streaming state
219330Opus 5.513/16—
Read a pattern, then detect it in a second stream. The pattern is supplied at runtime rather than built into the program.Hard
Streaming state Memory & interpreters
166209Gemini 4 Argon12/1616
Find the missing value in an unordered sequence of otherwise consecutive integers. A summary of the sequence can replace sorting it.Hard
Sequence algorithms Arithmetic construction
105293GPT-6.1 Sol13/162
Convert a decimal value to its octal notation. Repeated base-8 decomposition must be reassembled into the output representation.Medium
Arithmetic construction Memory & interpreters
136171Opus 5.516/167
Sort longer sequences terminated by −1. Buffer capacity and movement between memory nodes become central constraints.Hard
Sequence algorithms Memory & interpreters
338903Opus 5.59/1610
Emit an input's prime factors in ascending order as a zero-terminated sequence. Trial division must preserve the remaining quotient.Hard
Arithmetic construction Sequence algorithms
4091,110Opus 5.512/1614
Signal Exponentiator NEXUS.21.601.6
Raise one input to the power of another without multiplication or exponentiation instructions. Nested arithmetic magnifies loop costs.
Hard
Arithmetic construction
2891,259Opus 5.516/161
T20 Node Emulator NEXUS.22.280.8
Interpret commands for a tiny machine with two stored values: load, swap, add, and output. Command dispatch must preserve machine state.
Medium
Memory & interpreters Streaming state
168182Gemini 4 Argon12/16—
T31 Node Emulator NEXUS.23.727.9
Emulate eight addressable memory slots using read and write commands. Stack memory must be organized to support indexed access.
Hard
Memory & interpreters
151215Opus 5.515/164
Accumulate four input streams independently and output the index with the greatest running total. Ties choose the lower index.Hard
Streaming state Routing & synchronization
245282Gemini 4 Argon16/167

Community tournament puzzles 8

PuzzleTierMechanismsCommunitycyclesBest LLMcyclesSolvedmodelsNewpoints
Identify which of three repeating numeric sequences a subsequence belongs to, then predict its next value.Hard
Streaming state Sequence algorithms
177192GPT-6 Astra7/16—
Output the thousands digit of a power. The intermediate number can exceed the machine's numeric range, so direct evaluation is problematic.Hard
Arithmetic construction
41,39742,133Opus 5.52/16—
Striation Plotter SPEC35049542
Increment the color of a requested image row, wrapping after color 4. Each row has its own persistent color state.
Hard
Image generation Streaming state
1,9552,818Fable 5.18/16—
Round a value to the nearest multiples of N and 11−N, with ties rounded down. Two related outputs must share limited arithmetic resources.Hard
Arithmetic construction Routing & synchronization
2,694★ 1,918Opus 5.512/1615
Numeral Amalgamator SPEC49083185
Extract specified digits from three input numbers, then concatenate those digits into one output number.
Hard
Arithmetic construction Routing & synchronization
596898Opus 5.515/1614
Find values present in all three zero-terminated sequences, preserving their order in each sequence's output.Hard
Sequence algorithms Memory & interpreters
3,872★ 3,460GPT-6.1 Sol9/166
Reproduce the tournament's fixed Hwair image. The challenge is encoding the target pattern efficiently across the available nodes.Hard
Image generation
7281,120GPT-6.1 Sol11/16—
Error: 8 SPEC79324720
A garbled specification describes a counter that advances by 8 per input pair. Emit the upper bound if the counter is above it, the lower bound if below it, and zero otherwise.
Hard
Streaming state Arithmetic construction
198★ 186Gemini 4 Argon11/1612