Orak - Street Fighter III: leaderboard
Orak Street Fighter III: an LLM agent fights the game's built-in bot from a text game state (health, gauges, distance) using 15–20 discrete moves; score is stages cleared, normalized to 0–100 and averaged over trials, with the benchmark's default agentic strategy for the game. Higher is better. From KRAFTON AI's Orak benchmark (NeurIPS 2025).
Metric: Game Score (0-100). Source: krafton-ai.github.io. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 Mini (2025-01-31) | 33.3 | #249 |
| 2 | GPT-4o (2024-08-06) | 29.7 | #343 |
| 3 | DeepSeek R1 | 20 | #262 |
| 4 | Qwen 2.5 3B Instruct | 20 | #1159 |
| 5 | GPT-4o Mini | 16.7 | #598 |
| 6 | Qwen 2.5 7B Instruct | 16.7 | #863 |
| 7 | Claude 3.7 Sonnet | 16.7 | #259 |
| 8 | Llama 3.2 3B Instruct | 13.3 | #1327 |
| 9 | Gemini 2.5 Pro (Preview 03-25) | 13.3 | #126 |
| 10 | Llama 3.2 1B Instruct | 0 | #1518 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=orak-street-fighter-iii · How It Works · Data refreshed daily, snapshot 2026-11-05.