Playground · Simulation · HTML · Standing test
Conway's Game of Life
Conway's Game of Life has two rules, so there is nowhere to hide: one wrong neighbour count and a blinker stops blinking. We asked for a grid of at least 60 by 40, click to toggle, play, pause, step, clear, a speed control and three classic patterns. Then we cleared each board, clicked in a blinker and stepped it, drew a glider and stepped it four times, and dropped a Gosper glider gun to see whether it fires. Watch the gun: that is where the answers split.
The prompt, word for word
Make Conway's Game of Life: a grid of at least 60 by 40 cells, click cells to toggle them, buttons to play, pause, step one generation and clear, a speed control, and a few classic patterns (glider, pulsar, Gosper glider gun) to drop in. It must follow the exact rules of the game. Make it a single self-contained HTML file (inline CSS and JavaScript, no external libraries, fonts or images).
Every answer runs live, in a sandboxed frame. Click into one to play with it.
How each model did
| Model | Our notes | Checks | Time | Output tokens | Cost |
|---|---|---|---|---|---|
| GPT-6.1 Sol | A plain dark-blue board of 80 by 48 that starts empty, with a row of tools (Draw / erase, Glider, Pulsar, Gosper glider gun) and a blue preview before you drop a pattern. The blinker flips, the glider moves one cell diagonally in four steps, and the gun fires a steady stream of gliders. It does everything asked and nothing more. | 4 of 4 | 79 s | 5,088 450 thinking |
$0.051 |
| Claude Opus 5.5 Our pick | An 80 by 50 board that opens with a gun already running, plus LWSS, R-pentomino, a rotate button and a wrap-edges switch alongside the three asked-for patterns. Every test passed: the blinker flips, the drawn glider keeps its shape and moves one cell diagonally, and the dropped gun fires gliders down and to the right. The most complete of the correct answers, with a pattern that stays armed so you can stamp several in a row. | 4 of 4 | 70 s | 8,631 1,577 thinking |
$0.173 |
| Claude Fable 5.1 | A compact 80 by 50 board with a pattern dropdown, a "Place at center" button, a wrap toggle (on by default) and keyboard shortcuts listed under the toolbar. The blinker, glider, pulsar and gun all behave correctly, and the gun had fired two gliders after 85 generations. As correct as Opus, but at $0.51 it is the most expensive answer on the page. | 4 of 4 | 104 s | 10,127 2,368 thinking |
$0.508 |
| Gemini 3.8 Flash | An 80 by 48 board in a framed panel, with the controls underneath and Glider, Pulsar, Gosper Gun and LWSS stamp buttons. All the rules tests passed, and the gun fires gliders that wrap round to the top left of the board. The speed slider is labelled in "fps" rather than generations, a small oddity, and it was the fastest answer at 56 seconds. | 4 of 4 | 56 s | 9,751 3,408 thinking |
$0.037 |
| Muse Spark 1.3 | A tidy board of exactly 60 by 40 with cyan cells, a dead border and buttons that drop each pattern in the centre. The rules are right (the blinker flips, the glider glides, the pulsar repeats every three steps), but the Gosper gun is drawn wrongly: its left half is a ring, not the real pattern, and after 90 generations it has churned into debris without firing a single glider.
|
3.5 of 4 | 99 s | 8,210 3,220 thinking |
$0.035 |
| Grok 4.7 | A cream, paper-like page with a serif title and dark cells on an 80 by 50 grid, with pill-shaped buttons and the rules spelled out ("Birth 3 · Survival 2 or 3 · Edges stay dead"). Every test passed and the gun fires gliders cleanly. The only light-themed answer, and it took 182 seconds. | 4 of 4 | 182 s | 12,921 7,993 thinking |
$0.078 |
| Kimi K3 | A 90 by 60 board, the largest grid here, with the rules printed under a monospace title and a tool row for Draw / Toggle, Glider, Pulsar and Gosper Glider Gun. The blinker, glider, pulsar and gun all behave correctly, and the gun fires a clean stream of gliders. Patterns are centred on the click and clipped at the edge, so dropping the gun near a corner cuts part of it off. | 4 of 4 | 66 s | 10,968 5,882 thinking |
$0.143 |
| Qwen3.8 Max (0902) | The most styled answer: a 90 by 56 "cellular automaton" console with counters, undo, a torus or dead-border switch and pattern cards with small previews. The rules are right and the glider and pulsar work, but the Gosper gun has a 2 by 3 block where the 2 by 2 block should be (38 cells instead of 36), so it explodes into chaos instead of firing gliders. It also took 859 seconds, the slowest answer by far.
|
3.5 of 4 | 859 s | 34,953 22,687 thinking |
$0.210 |
| DeepSeek V4.1 Flash | A monospace 80 by 50 board with a pattern dropdown that also includes LWSS, Pentadecathlon, R-pentomino and Acorn, and edges that wrap. Every test passed and the gun fires gliders. The cheapest fully correct answer at $0.021, though it took 173 seconds. | 4 of 4 | 173 s | 17,442 11,284 thinking |
$0.021 |
| Mistral Large 4 | The first two tries ran out of tokens while still thinking; the third, with reasoning turned off, gave an 80 by 50 board with red cells and a dropdown of twelve patterns. The rules are right (the blinker flips and the glider glides), but the pulsar comes out as nine 3 by 3 blocks (84 cells, not 48) and never repeats, and the gun is missing its right-hand block, so it falls apart without firing gliders. A chosen pattern also stays selected, so later clicks keep stamping it instead of toggling cells.Took 3 tries: ran out of room at 64,000 tokens; ran out of room at 128,000 tokens; finished at 128,000 tokens, thinking off.
|
3 of 4 | 112 s | 5,971 | $0.013 $0.414 all tries |
One answer per model, at each model's default settings unless the card says otherwise, through OpenRouter. If an answer ran out of room or stopped short, we asked again with more room, then with thinking off, and say so. We tried every answer, with Claude Opus 5.5 running the checks and drafting the notes; it is an Anthropic model, so we check its calls on Claude answers ourselves. This is a quick look, not a benchmark: run it again and every answer will differ. For measured results, see our benchmarks.
What we checked, by trying each answer ourselves: Follows the rules exactly (blinker flips, glider glides); Play, pause, step, clear and speed work; Glider, pulsar and Gosper glider gun drop in and behave correctly; Grid at least 60 by 40. A partial pass counts as half.
Verdict
Every model got the rules exactly right: in all ten, the blinker flipped and the drawn glider moved one cell diagonally in four steps, and play, pause, step, clear and speed all worked. The difference is the patterns. Claude Opus 5.5 is our pick: everything is correct, and it adds rotate, extra patterns and a wrap switch, in 70 seconds for $0.17. Claude Fable 5.1, GPT-6.1 Sol, Gemini 3.8 Flash, Grok 4.7, Kimi K3 and DeepSeek V4.1 Flash are also fully correct. Gemini was the fastest at 56 seconds for $0.037, and DeepSeek the cheapest at $0.021. Fable cost $0.51 for much the same result.
Three answers got the Gosper gun wrong: Muse Spark's and Qwen's guns turn into debris instead of firing, and Mistral's pulsar and gun are both malformed. Qwen built the most elaborate interface but took over 14 minutes. Mistral needed three tries: the first two ran out of tokens while thinking (64,000 and then 128,000).
What would your prompt get?The Spring Prompt playground runs one prompt on up to six models from OpenAI, Anthropic, Google, Mistral and others, side by side.
Ask for access
More from the playground
