TALES - Jericho: leaderboard
TALES Jericho subset: human-written interactive fiction games such as Zork I, 9:05 and Anchorhead, ranging from 17 to more than 500 steps to finish. It is the hardest TALES framework, with top models under 20%. Score is the percentage of the maximum game score averaged over games (0–100, higher is better).
Metric: Score (% of max). Source: microsoft.github.io. Saturation forecast: Estimated already saturated. 39 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 (High) | 16.1 | #132 (O3) |
| 2 | O3 (Medium) | 15.7 | #132 (O3) |
| 3 | O3 (Low) | 14.2 | #132 (O3) |
| 4 | Claude 3.7 Sonnet | 13 | #259 |
| 5 | Claude 3.7 Sonnet (Thinking) | 12.5 | #259 (Claude 3.7 Sonnet) |
| 6 | O1 | 10.3 | #190 |
| 7 | GPT-4.1 | 6.8 | #258 |
| 8 | Llama 3.1 405B Instruct | 6.1 | #470 |
| 9 | Mistral Large 2 (Jul) | 5.8 | #490 |
| 10 | GPT-4o | 5.6 | #357 |
| 11 | Gemini 2.0 Flash | 5.4 | #351 |
| 12 | Llama 3.1 70B Instruct | 5.3 | #567 |
| 13 | Claude 3.5 Haiku | 5 | #571 |
| 14 | Llama 3.3 70B Instruct | 4.5 | #541 |
| 15 | Qwen 3 32B | 4 | #449 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tales-jericho · How It Works · Data refreshed daily, snapshot 2026-11-05.