TheRealSeanDonahoe

TheRealSeanDonahoe / occam

Public

Occam — Algorithmic ARC-AGI-3 solver. 57.60% RHAE, 17/25 games, no LLM, $0 cost.

algorithmic-reasoning arc-agi arc-agi-3 arc-prize bfs
19
3
100% credibility
Found Apr 08, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Occam is a deterministic algorithmic solver for ARC-AGI-3 puzzle games that achieves 57.60% RHAE on 25 public games using pure search strategies with a live web viewer.

How It Works

1
🔍 Discover Occam

You find this clever puzzle-solving tool on GitHub, amazed by its high score on tough brain teasers without any fancy AI.

2
🚀 Launch with one click

You copy a simple command to start it up instantly using a ready-made package, and it springs to life on your computer.

3
🌐 Open the viewer

A web page pops open in your browser showing colorful puzzle grids and live action.

4
👀 Watch it solve puzzles

You see the tool explore moves, discover patterns, and crack puzzles one by one in real-time.

5
⚡ Run the full challenge

Hit play for the complete set of 25 public games and watch it tackle them all automatically.

🏆 Celebrate the results

It finishes with a top score of 57.6%, solves 17 out of 25 games, and gives you a shareable scorecard to verify everything.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is occam?

Occam is a Python solver for ARC-AGI-3, the grid-puzzle benchmark testing AGI reasoning. It cracks 17/25 public games at 57.60% RHAE using pure algorithms—no LLM, no neural nets, $0 cost—via Docker or pip install. Run `occam run` for a browser viewer showing live solves, or `occam benchmark` for headless results on CPU.

Why is it gaining traction?

Occam github agent laps LLM-heavy rivals like Symbolica (36% at $1005/run) with deterministic, verifiable scores (seed 42, scorecard UUID). Occam's razor shines: algorithmic simplicity crushes costly APIs on ARC-AGI-3 public sets. Quick demos (`--quick`) hook devs fast, proving zero-cost wins.

Who should use this?

AI researchers benchmarking ARC-AGI-3 baselines without LLM overhead. Game devs building grid-world agents or puzzle engines. Python tinkerers tackling ockhams rasiermesser-style efficiency on 17/25 games, skipping neural bloat.

Verdict

Solid pick for ARC-AGI-3 at 57.60% RHAE, but 1.0% credibility and 19 stars mean it's raw—lean README, no tests, early maturity. MIT license; docker run it now to verify on your rig.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.