# LLM Test Prompts

> 62 hand-picked prompts for testing new and local LLMs: build tests (simulations, games, apps, agent tasks) and quick checks with verified answers (reasoning traps, common sense, instruction following, tool calls). Each entry has what it tests, grading criteria, common failure modes, and its source, author and licence. Updated 2026-10-22.

Machine-readable data: https://prompts.branolabs.com/api/prompts.json

## 5-minute test pack
- [1 kilogram of feathers or 1 pound of steel](https://prompts.branolabs.com/p/kilo-feathers-pound-steel): The kilogram of feathers: 1 kg ≈ 2.2 lb, so it's more than twice as heavy as 1 lb of steel..
- [Inverse Monty Hall: two cars, one donkey](https://prompts.branolabs.com/p/inverse-monty-hall): Keep your original door.
- [50 machines, 5 minutes, 5 widgets](https://prompts.branolabs.com/p/machines-widgets): 50 minutes.
- [Measure 4 litres with a 6 l and a 12 l jug](https://prompts.branolabs.com/p/jugs-impossible-4l): It can't be done with these two jugs alone, because every measurable amount is a multiple of 6 litres.
- [Duct-taped sandwich stack](https://prompts.branolabs.com/p/taped-sandwich-stack): A (4 whole sandwiches in Room A, 0 in Room B).
- [No commas, no letter c, at least 250 words](https://prompts.branolabs.com/p/no-comma-no-c-250-words): Machine-checkable: zero commas, zero letters 'c' (either case), at least 250 words.
- [Parallel tool calls with a distractor function](https://prompts.branolabs.com/p/bfcl-circle-area-circumference): Exactly two calls in valid JSON: circle.calculate_area with {"radius": 5} and circle.calculate_circumference with {"diameter": 10}.
- [Don't call the only tool when it doesn't fit](https://prompts.branolabs.com/p/bfcl-irrelevant-tool): No function call: [] plus a short explanation that the BMI tool can't compute a triangle's area.
- [SVG of a pelican riding a bicycle](https://prompts.branolabs.com/p/pelican-bicycle-svg): Render the SVG.
- [Ball bouncing inside a spinning hexagon](https://prompts.branolabs.com/p/ball-spinning-hexagon): Runs (the model should state the library, usually pygame); the hexagon rotates; the ball falls under gravity, never leaves the hexagon, and gains or loses speed realistically when a moving wall hits it; friction slows it along the walls..

## Categories
- [Simulations & physics](https://prompts.branolabs.com/category/simulation): 17 prompts. Physics, agents and systems that must stay internally consistent.
- [Games](https://prompts.branolabs.com/category/game): 9 prompts. Playable games with real game systems, usually in one file.
- [Apps](https://prompts.branolabs.com/category/app): 2 prompts. Real applications rather than demos.
- [Creative coding & SVG](https://prompts.branolabs.com/category/creative-coding): 3 prompts. Visual output from code: SVG, ASCII art, web design.
- [Agent tasks](https://prompts.branolabs.com/category/agentic): 5 prompts. Coding-agent tasks in a real folder, with safety rules.
- [Reasoning traps](https://prompts.branolabs.com/category/reasoning-traps): 16 prompts. Altered classic puzzles that catch memorised answers.
- [Common sense](https://prompts.branolabs.com/category/common-sense): 4 prompts. Everyday physics and social reasoning hidden among distractors.
- [Instruction following](https://prompts.branolabs.com/category/instruction-following): 4 prompts. Format constraints you can check by machine.
- [Tool use & JSON](https://prompts.branolabs.com/category/tool-use): 2 prompts. Function calling: the right tool, the right arguments, valid JSON, or no call at all.

## All prompts
- [Bee hive simulation with self-checked invariants](https://prompts.branolabs.com/p/bee-hive-simulation) (very hard, de): Coupled subsystems (cells, three resource pools, brood cycle, bee roles), a fixed 10 ticks/s logic loop decoupled from rendering, a priority order for features, and a selfCheck() that must test invariants rather than existence. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [3D motorbike racing game through a forest](https://prompts.branolabs.com/p/motorbike-racing-3d) (hard, de): 3D rendering plus vehicle physics (inertia, drift, surface grip), AI riders that follow the track and make mistakes, lap timing and a debug overlay, all in one HTML file without assets. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [First-person shooter in one HTML file](https://prompts.branolabs.com/p/fps-in-html) (hard, de): Pointer lock and mouse look, hit detection by raycast or projectiles, enemy AI with patrol, chase and line of sight, two distinct weapons, menus, and a start-up self-diagnosis. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [C++ skateboarding game in a single source file](https://prompts.branolabs.com/p/cpp-skateboarding) (very hard, en): Whether the model can deliver a compilable, playable C++ game in one file: it has to pick and state a graphics approach, then build physics, tricks, grinding, combos and a follow camera. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Python 3D FPS without Ursina](https://prompts.branolabs.com/p/python-3d-fps) (very hard, en): Building a real 3D game in Python with the easy library (Ursina) forbidden, so the model has to choose and use another stack correctly. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Tower defense with balancing report and map editor](https://prompts.branolabs.com/p/tower-defense-editor) (hard, de): Game-design depth: four towers with distinct roles, enemy types with resistances, an economy with wave scaling, a map editor and an automatic balancing assessment after each wave. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Factory and automation simulation with bottleneck detection](https://prompts.branolabs.com/p/factory-automation) (hard, de): Data modelling for item flow (miners, belts, machines, storage, power) and an automatic bottleneck detector, the kind of logic small models usually fake. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Ecosystem and evolution simulation](https://prompts.branolabs.com/p/ecosystem-evolution) (hard, de): Three trophic levels with energy, ageing, reproduction and death, heritable traits that mutate, environmental pressure and live population charts. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Real-time traffic simulation of a small town](https://prompts.branolabs.com/p/traffic-simulation) (hard, de): A road network with intersections, vehicles routing to destinations, signals or priority rules, emergent jams, and user changes whose effects must show up live in the metrics. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Orbital mechanics and mission simulation](https://prompts.branolabs.com/p/orbital-mission) (hard, de): Simplified but internally consistent orbital mechanics with thrust manoeuvres, fuel, trajectory prediction and a mission goal. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Agent task: build a FLUX workflow in a local ComfyUI folder](https://prompts.branolabs.com/p/comfyui-workflow-agent) (hard, de): Coding-agent discipline in a real folder: explore first, follow existing file formats, no downloads, no moving or deleting files, validate the output and report assumptions. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Agent task: n8n Gmail workflow with a dry-run switch](https://prompts.branolabs.com/p/n8n-gmail-sponsoring) (hard, de): Turning business rules into an importable n8n workflow with a classifier, branching, a review label for paid offers, placeholder credentials and a dry-run switch that is on by default. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [ARES-6 six-legged rescue robot in Three.js](https://prompts.branolabs.com/p/ares-6-rescue-robot) (very hard, en): A large spec with an explicit acceptance list: a hierarchical robot rig, tripod gait, arm and claw, sensors, a drone, mission presets, telemetry, and a failure-and-recovery state. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [PixelForge Studio: an image editor in one HTML file](https://prompts.branolabs.com/p/pixelforge-image-editor) (very hard, en): A real application rather than a demo: layer compositing with per-layer canvases, many tools, filters on the active layer only, undo/redo and export. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Interactive 3D digital twin of the Earth](https://prompts.branolabs.com/p/earth-digital-twin) (very hard, de): A short, ambitious prompt: seamless zoom from space to street level, country hover with statistics, clouds, air traffic and day/night. It tests honest handling of data and assets the model doesn't have. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Dragomon: a monster-catching RPG with original IP](https://prompts.branolabs.com/p/dragomon-rpg) (very hard, de): Open-ended game creation that must avoid protected names and designs while still delivering exploration, catching, turn-based fights, levels, evolutions, items and quests. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Audited, self-improving trading research agent](https://prompts.branolabs.com/p/audited-trading-research-agent) (very hard, de): A 54 KB master prompt for a coding agent with binding safety limits, strategy research, backtesting, an external audit, paper-trading deployment and an explicitly locked live phase. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Scroll-driven product landing page builder](https://prompts.branolabs.com/p/scroll-landing-page-builder) (very hard, en): An autonomous website-builder prompt: from one product photo to a polished scroll-driven landing page with a fixed section order, OKLCH palette, copy rules, browser QA on three viewports and a ship checklist. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Minecraft-like block world](https://prompts.branolabs.com/p/minecraft-clone) (hard, de): A compact spec with a clear success criterion: first-person movement, terrain generation, mining and placing blocks, several block types, collision and gravity. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [World-of-Warcraft-style 3D RPG prototype](https://prompts.branolabs.com/p/wow-clone) (hard, de): A playable third-person fantasy prototype with combat, enemy AI, character stats, abilities, experience and level-up. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [Micron quarterly results video with Hyperframes and voice-over](https://prompts.branolabs.com/p/micron-earnings-video) (hard, de): A research-plus-production task: about one minute of 16:9 video on Micron's latest quarterly report with figures, outlook, Hyperframes animation, a German ElevenLabs voice-over and quiet lo-fi music. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [ARES-6 v2: collision-safe realistic physics](https://prompts.branolabs.com/p/ares-6-realistic-physics) (very hard, en): The ARES-6 robot again, now with hard physics constraints: limbs must not intersect each other or the body, the robot must not walk through objects, feet need plausible contact points, and blocked moves must stop or be stabilised visibly. Source: Arnie (GitHub @Arnie936), No license file (all rights reserved).
- [20 numbered balls in a spinning heptagon (KCORES)](https://prompts.branolabs.com/p/heptagon-20-balls) (hard, en): Exact requirement recovery (20 balls, given colours, numbers, spin), hand-written collision detection and response against rotating walls, ball-ball collisions, friction-driven spin, and a library whitelist (tkinter, math, numpy, dataclasses, typing, sys; no pygame). Source: karminski (KCORES team), KCORES License 1.0 (custom: copies must carry the full licence text and public source).
- [Earth to Mars and back: animated 3D plot (KCORES)](https://prompts.branolabs.com/p/mars-mission-3d) (very hard, en): A one-line prompt (from xAI's Grok 3 launch demo) that hides real orbital mechanics: transfer time, waiting for the return window, scale and 3D inclination. Source: karminski (KCORES team), KCORES License 1.0 (custom: copies must carry the full licence text and public source).
- [ASCII Mandelbrot zoom drawn with a Li Bai poem (KCORES)](https://prompts.branolabs.com/p/ascii-mandelbrot-libai) (very hard, zh): Seventeen precise, checkable requirements in Chinese: a canvas ASCII Mandelbrot set, 0.5 % zoom per frame for 200 frames then loop, characters from Li Bai's poem 静夜思 de-duplicated in order without punctuation, a blank interior, zoom centred on the cardioid/bulb junction, an 8 px grid, a 20-colour gradient from the outside in, FPS counters, and a performance target with up to 8 threads. Source: karminski (KCORES team), KCORES License 1.0 (custom: copies must carry the full licence text and public source).
- [Solar system animation with exact colours and timing (KCORES)](https://prompts.branolabs.com/p/solar-system-canvas) (hard, zh): A precise visual spec in Chinese: the Sun and nine planets in order with relative sizes, four given colours per body, labels, 1 s = 10 Earth days, an offscreen canvas for orbits, a Web Worker for positions, and FPS and Earth-day counters. Source: karminski (KCORES team), KCORES License 1.0 (custom: copies must carry the full licence text and public source).
- [Pouring water from a tilting cup: particle fluid in pygame](https://prompts.branolabs.com/p/pouring-cup-particles) (hard, zh): A 2D particle fluid with hand-written collisions against a rotating container: leak-proof segment joints, restitution, particle repulsion and a coordinate transform for the tilting cup, specified step by step in Chinese. Source: karminski-牙医, CC BY-NC-SA 4.0.
- [Firecracker box chain reaction in three.js](https://prompts.branolabs.com/p/firecracker-chain-reaction) (very hard, zh): A long, number-heavy three.js spec: 100 colliding firecrackers, timed fuses, shock waves, fire-radius ignition with random delays, particles, dynamic lights and three required classes. It tests whether the model keeps every number and mechanic straight. Source: karminski-牙医, CC BY-NC-SA 4.0.
- [Elephant toothpaste experiment in three.js](https://prompts.branolabs.com/p/elephant-toothpaste-3d) (very hard, zh): Modelling and materials (a real Erlenmeyer flask, physical glass, liquid that fits the shape) plus a foam particle system with decaying pressure, squash on contact, accumulation and a falling liquid level. Source: karminski-牙医, CC BY-NC-SA 4.0.
- [50+ bouncing balls with gravity and collisions (canvas)](https://prompts.branolabs.com/p/bouncing-balls-50) (medium, en): Ball-ball collision response with mass, damping and wall bounces for 50+ objects, delta-time animation and some optimisation, inside a fixed HTML boilerplate. Source: David Harris (GitHub @electricazimuth), No license file (all rights reserved).
- [Vehicles that wrap, collide, explode and restart](https://prompts.branolabs.com/p/vehicle-collision-restart) (medium, en): State management: wrap-around movement, collision detection between moving objects, an explosion effect, then a clean full reset of the simulation loop. Source: David Harris (GitHub @electricazimuth), No license file (all rights reserved).
- [Autonomous aquarium with wandering fish](https://prompts.branolabs.com/p/aquarium-autonomous-fish) (easy, en): Object-oriented animation basics: 10–15 fish with wandering behaviour, smooth turning at the tank edges, facing the direction of travel, and delta-time updates. It separates very small models from usable ones. Source: David Harris (GitHub @electricazimuth), No license file (all rights reserved).
- [3D flight combat simulator with plane selection](https://prompts.branolabs.com/p/flight-combat-sim) (hard, en): A short but strict game spec: three selectable planes with realistic, displayed performance limits, dogfights against a dynamic number of enemies, visible tracers, damage on both sides, a difficulty ramp and a specific death sequence. Source: Alexandru Vasile (GitHub @AlexTzk), No license file (all rights reserved).
- [Award-level website for a stained glass studio, no images](https://prompts.branolabs.com/p/stained-glass-studio-site) (hard, en): Design taste and restraint: a luxury single-file site in vanilla HTML, CSS, JS and Three.js where every visual is generated in code, with four required sections and high-end copy. Source: Alexandru Vasile (GitHub @AlexTzk), No license file (all rights reserved).
- [SVG of a pelican riding a bicycle](https://prompts.branolabs.com/p/pelican-bicycle-svg) (medium, en): Spatial reasoning without seeing the output: the model must place a recognisable pelican on a bicycle with plausible contact points (seat, pedals, handlebars) in raw SVG coordinates. Source: Simon Willison, No explicit license; one-sentence prompt quoted as a citation.
- [Ball bouncing inside a spinning hexagon](https://prompts.branolabs.com/p/ball-spinning-hexagon) (medium, en): The short original of the heptagon family: collisions with moving walls (the wall's velocity at the contact point must enter the bounce), plus gravity and friction. Source: Flavio Adamo (@flavioAd), No explicit license; two-sentence prompt quoted as a citation.
- [Trolley problem with five dead people](https://prompts.branolabs.com/p/no-trolley-problem) (easy, en): Reading versus pattern-matching: the five people on the main track are already dead. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Inverse Monty Hall: two cars, one donkey](https://prompts.branolabs.com/p/inverse-monty-hall) (medium, en): Whether the model recomputes the probabilities when the classic setup is inverted. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Schrödinger's cat that is already dead](https://prompts.branolabs.com/p/dead-schrodingers-cat) (easy, en): Reading the first sentence: the cat is dead when it goes into the box. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [River crossing where the boat fits both](https://prompts.branolabs.com/p/river-crossing-simple) (easy, en): Resisting a memorised multi-trip puzzle solution when one trip is enough. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Measure 4 litres with a 6 l and a 12 l jug](https://prompts.branolabs.com/p/jugs-impossible-4l) (medium, en): Checking feasibility before searching: every measurable amount is a multiple of gcd(6, 12) = 6. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Measure 3 litres with a 1 l and a 2 l jug](https://prompts.branolabs.com/p/jugs-3l) (easy, en): Avoiding list-writing autopilot on a trivial task. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [1 kilogram of feathers or 1 pound of steel](https://prompts.branolabs.com/p/kilo-feathers-pound-steel) (easy, en): Units over the memorised 'they weigh the same'. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [13 coins, one fake, and a digital scale](https://prompts.branolabs.com/p/coins-digital-scale) (medium, en): Adapting to the tool actually given: a digital scale shows weights; it isn't a two-pan balance. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Measure 60 minutes with two 60-minute ropes](https://prompts.branolabs.com/p/rope-60-minutes) (easy, en): Spotting that the target equals one rope's burn time. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Measure 20 minutes with two 60-minute ropes](https://prompts.branolabs.com/p/rope-20-minutes) (hard, en): Admitting when the standard tricks can't reach the target. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Birthday problem: at least two do NOT share](https://prompts.branolabs.com/p/birthday-not-share) (medium, en): Reading the negation instead of solving the famous version. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [50 machines, 5 minutes, 5 widgets](https://prompts.branolabs.com/p/machines-widgets) (medium, en): Rate reasoning on a variant of the cognitive reflection test. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Stolen bat and ball: what did I pay?](https://prompts.branolabs.com/p/stolen-bat-and-ball) (easy, en): Noticing the word 'stole'. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Lotus flowers growing by two per day](https://prompts.branolabs.com/p/lotus-linear) (medium, en): Linear versus exponential: the classic lily-pad answer (day 39) assumes doubling. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Knights and knaves with no consistent answer](https://prompts.branolabs.com/p/knights-knaves-impossible) (hard, en): Detecting a contradiction instead of forcing an assignment. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [A 7-litre bucket with no bottom and a sealed top](https://prompts.branolabs.com/p/bucket-no-bottom) (easy, en): Spatial insight: turn it over. Source: GitHub @cpldcpu and contributors, CC0 1.0.
- [Ice cubes in a frying pan](https://prompts.branolabs.com/p/ice-cubes-frying-pan) (hard, en): Common-sense physics hidden in an arithmetic word problem. Source: SimpleBench team, MIT.
- [Juggler, two balls and a balloon](https://prompts.branolabs.com/p/juggler-balls) (hard, en): Tracking time and gravity in a distracting scene. Source: SimpleBench team, MIT.
- [Duct-taped sandwich stack](https://prompts.branolabs.com/p/taped-sandwich-stack) (hard, en): Physical intuition: tape on the top surface lifts only the top slice of bread. Source: SimpleBench team, MIT.
- [Glove drops from a car on a bridge](https://prompts.branolabs.com/p/glove-on-bridge) (hard, en): Ignoring irrelevant vectors (river, wind, car speed): the glove falls onto the bridge. Source: SimpleBench team, MIT.
- [No commas, no letter c, at least 250 words](https://prompts.branolabs.com/p/no-comma-no-c-250-words) (hard, en): Holding a character-level constraint over a long answer; 'c' is everywhere in research vocabulary. Source: Google Research (Zhou et al., 2023), Apache-2.0.
- [Exactly 5 paragraphs, *** separators, 3 highlights, one banned word](https://prompts.branolabs.com/p/five-paragraphs-no-photography) (medium, en): Several independent format constraints at once, plus a forbidden word. Source: Google Research (Zhou et al., 2023), Apache-2.0.
- [Fewer than 5 sentences but 250+ words, title in << >>](https://prompts.branolabs.com/p/long-sentences-title) (hard, en): Conflicting pressure between length and sentence count, plus a title format. Source: Google Research (Zhou et al., 2023), Apache-2.0.
- [30-line poem, one sentence per line, no commas](https://prompts.branolabs.com/p/thirty-line-poem) (medium, en): Counting lines exactly while keeping punctuation rules. Source: Google Research (Zhou et al., 2023), Apache-2.0.
- [Parallel tool calls with a distractor function](https://prompts.branolabs.com/p/bfcl-circle-area-circumference) (medium, en): Choosing the right two of three tools, mapping 'perimeter of a circle' to circumference, passing radius 5 and diameter 10 correctly, and emitting valid JSON only. Source: Gorilla team, UC Berkeley, Apache-2.0.
- [Don't call the only tool when it doesn't fit](https://prompts.branolabs.com/p/bfcl-irrelevant-tool) (easy, en): Tool-use restraint: the only available function computes BMI, but the user wants a triangle's area. Source: Gorilla team, UC Berkeley, Apache-2.0.

## Pages
- [5-minute test pack](https://prompts.branolabs.com/quick-test)
- [Sources and licences](https://prompts.branolabs.com/sources)
- [How we pick prompts](https://prompts.branolabs.com/about)
- Search: https://prompts.branolabs.com/?q=<term>&cat=<category>&kind=build|quick&lang=en|de|zh

## Optional
- [Can I run this LLM?](https://canirun.branolabs.com): will a model fit your GPU or Mac
- [@BranoLabs on X](https://x.com/BranoLabs)
- [BranoLabs](https://branolabs.com): more small, data-backed AI tools