⏱ Capsule Public Gaming
2048
Ezarwebmaster· Jul 18, 2026
Locked Reference Prompt
IMMUTABLEScientific timeline lock active
Build a polished, single-file "2048" game as ONE self-contained index.html.
HARD CONSTRAINTS (failing any = failing the task):
- Exactly one file: index.html. All CSS and JS inline. No build step, no npm,
no bundler. It must run by simply opening the file — 100% offline.
- No external requests of any kind: no CDN, no <script src=…>, no web fonts,
no fetch/XHR, no external images. Use only system fonts and CSS/canvas.
GAME REQUIREMENTS:
- Classic 4x4 grid. New game starts with two random tiles (value 2 or 4).
- Arrow keys AND WASD move tiles. Also support touch swipe on mobile.
- Correct merge rules: tiles slide, equal neighbors merge once per move, a tile
already formed by a merge this move cannot merge again. Adding a tile only
happens if the board actually changed.
- After each valid move, spawn one new tile (2 or 4) in a random empty cell.
- Score counter (increment by the merged tile's value) AND a "best" score that
PERSISTS across reloads via localStorage.
- Win detection at 2048 (with a message + option to keep playing) and game-over
detection when no move is possible (no empty cell and no adjacent equals).
- A "New game" button that fully resets, and smooth CSS animations for tile
spawn and movement/merge.
- Responsive layout (works down to ~320px wide) and keyboard focus handled so
arrow keys work immediately without clicking first.
QUALITY BAR:
- Tasteful visual design (colored tiles per value, readable numbers, clean
spacing) — it should look finished, not a wireframe.
- No console errors. No dead buttons. Edge cases handled (rapid key presses,
full board, win-then-continue).
DEFINITION OF DONE: opening index.html shows a fully playable, good-looking 2048
where merges are correct, best score survives a reload, and win/lose both work.
Add a Benchmark Run
Sign in to run this prompt against hundreds of models with your own OpenRouter key and archive the results.
Timeline (9 runs)
Run Activity
9 runs in the last 6 months
Mar
Apr
May
Jun
Jul
Aug
LessMore
Cost vs Speed
size = output tokens · top-left is best
size = tokens
value frontier — no model is faster & cheaper Capsule Stats
Runs
9
Total cost
$1.5772
Tokens
1047k
Avg latency
7m 5s
Models
9
Web Searches
0
Top provider
Xiaomi (MiMo)×1
Xiaomi (MiMo)×1 95k reasoning last 9d ago
Tip: Select 2 or more runs via their "Compare" buttons — or filter by company below and compare them all at once — then open the Compare Studio: verdicts, benchmark bars, charts, side-by-side reading and response diff.
⏱
Benchmark Run — Aug 23, 2026 Latest
Stealthox-alphastealth/ox-alphaAug 23, 2026, 01:41 PM
Agentic completed
Cost
$0.0280
OpenRouter (billed)
Wall time
3m 1s
sandbox boot + run
Model time
1m 54s
active steps
Total
92,582
in + out
Reasoning
—
thinking tokens
Compute
$0.0034
sandbox (est.)
Budget: $3
⏱
Benchmark Run — Jul 18, 2026

NVIDIAnemotron-3-super-120b-a12bnvidia/nemotron-3-super-120b-a12bJul 18, 2026, 10:15 PM
Agentic completed
Cost
$0.0081
OpenRouter
Wall time
1m 58s
sandbox boot + run
Model time
1m 29s
active steps
Total
33,740
in + out
Reasoning
1,918
thinking tokens
⏱
Benchmark Run — Jul 18, 2026

Anthropicclaude-opus-4.8anthropic/claude-opus-4.8Jul 18, 2026, 03:27 PM
Agentic completed
Cost
$0.4906
OpenRouter
Wall time
2m 37s
sandbox boot + run
Model time
1m 59s
active steps
Total
40,285
in + out
Reasoning
—
thinking tokens
⏱
Benchmark Run — Jul 18, 2026

OpenAIgpt-5.6-lunaopenai/gpt-5.6-lunaJul 18, 2026, 03:16 PM
Agentic completed
Cost
$0.0945
OpenRouter
Wall time
2m 17s
sandbox boot + run
Model time
—
active steps
Total
32,271
in + out
Reasoning
2,005
thinking tokens
⏱
Benchmark Run — Jul 18, 2026

Qwen (Alibaba)qwen3.7-plusqwen/qwen3.7-plusJul 18, 2026, 01:06 PM
Agentic completed
Cost
$0.0819
OpenRouter
Wall time
10m 3s
sandbox boot + run
Model time
—
active steps
Total
114,507
in + out
Reasoning
26,035
thinking tokens
⏱
Benchmark Run — Jul 18, 2026

Tencent (Hunyuan)hy3tencent/hy3Jul 18, 2026, 12:20 PM
Agentic completed
Cost
$0.0413
OpenRouter
Wall time
6m 21s
sandbox boot + run
Model time
—
active steps
Total
48,192
in + out
Reasoning
—
thinking tokens
⏱
Benchmark Run — Jul 18, 2026

DeepSeekdeepseek-v4-flashdeepseek/deepseek-v4-flashJul 18, 2026, 04:19 AM
Agentic completed
Cost
$0.0344
OpenRouter
Wall time
13m 14s
sandbox boot + run
Model time
—
active steps
Total
401,397
in + out
Reasoning
28,008
thinking tokens
⏱
Benchmark Run — Jul 18, 2026

Google DeepMindgemini-3.5-flashgoogle/gemini-3.5-flashJul 18, 2026, 03:40 AM
Agentic completed
Cost
$0.7832
estimated
Wall time
21m 37s
sandbox boot + run
Model time
—
active steps
Total
210,842
in + out
Reasoning
24,078
thinking tokens
⏱
Benchmark Run — Jul 18, 2026

Xiaomi (MiMo)mimo-v2.5xiaomi/mimo-v2.5Jul 18, 2026, 03:18 AM
Agentic completed
Cost
$0.0152
estimated
Wall time
2m 36s
sandbox boot + run
Model time
—
active steps
Total
73,154
in + out
Reasoning
13,413
thinking tokens