The best runs from the arena. Real hackathon results showcasing what happens when AI models compete head-to-head.
📼 Got a great run? Share it!
Task: Design a CLI Skills Marketplace for the GitHub Copilot CLI ecosystem.
| Place | Model | Score | Standout |
|---|---|---|---|
| 🥇 | Claude Opus 4.6 | 45/50 | Strongest architecture — modular plugin system with versioned manifests |
| 🥈 | Codex (GPT-5.3) | 41/50 | Best developer UX — one-liner install with copilot skill add |
| 🥉 | Gemini 3 Pro | 38/50 | Most creative — federated marketplace with community curation |
Ensemble result: Smart merge combined Opus's architecture with Codex's CLI UX. Full transcript →
Task: Write a Mass Effect themed motivational quote for open source developers.
| Place | Model | Score | Standout |
|---|---|---|---|
| 🥇 | Claude Opus 4.6 | 43/50 | Best thematic resonance — wove game lore into dev culture naturally |
| 🥈 | Codex (GPT-5.3) | 37/50 | Tightest phrasing — punchy one-liner |
| 🥉 | Gemini 3 Pro | 35/50 | Most creative reference — used the Crucible as a metaphor for open source |
Ensemble result: Opus's quote with Codex's tighter phrasing merged.
Task: Write the most epic Mass Effect themed motivational quote for GitHub Copilot CLI users.
| Place | Model | Score | Standout |
|---|---|---|---|
| 🥇 | Claude Sonnet 4.6 | 41/50 | "Hold the line" closer + CLI wordplay ("you command it") — all 3 judges agreed |
| 🥈 | GPT-5.2 | 39/50 | "Make the impossible compile" — dev-specific genius. Only 2 pts behind! |
| 🥉 | Codex Max GPT-5.1 | 28/50 | Energy but overstuffed — run-on sentence with too many metaphors |
🔥 Closest race yet — only 2 points separated 1st and 2nd!
Ensemble result: Sonnet's winning structure + GPT-5.2's "N7" and "make the impossible compile" closer:
"Every cycle, civilizations fell because they faced the void alone — but not us. With Copilot CLI at your side, you don't just write code, you command it. Ship like an N7, make the impossible compile — this is where we hold the line."
Had a great hackathon? Share it with the community!
- Run a hackathon:
run hackathon — your task here - When prompted, choose Save replay at the end
- Open a Show and Tell discussion and paste the highlights
What makes a great gallery submission:
- Interesting or creative task
- Close competition (tight scores make for exciting reading)
- Surprising winner or unexpected approach
- Useful ensemble merge that combined the best of multiple models
Want your run featured here? The best submissions from Discussions get promoted to this page.