TALES - TextWorld Express: leaderboard
TALES TextWorld Express subset: text games from TextWorld Express, a fast re-implementation of TextWorld-style scenarios whose simplified parser does not accept near-miss action phrasings. Score is the percentage of the maximum game score averaged over games (0–100, higher is better), as reported on the TALES website.
Metric: Score (% of max). Source: microsoft.github.io. Saturation forecast: Estimated already saturated. 39 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 95.8 | #259 |
| 2 | GPT-4.1 | 92.5 | #258 |
| 3 | O3 (Medium) | 91.9 | #132 (O3) |
| 4 | Claude 3.7 Sonnet (Thinking) | 91.3 | #259 (Claude 3.7 Sonnet) |
| 5 | O3 (Low) | 89.8 | #132 (O3) |
| 6 | O3 (High) | 89.6 | #132 (O3) |
| 7 | Qwen 2.5 72B Instruct | 83.8 | #469 |
| 8 | Llama 3.1 70B Instruct | 81.9 | #567 |
| 9 | GPT-4o | 80.6 | #357 |
| 10 | Claude 3.5 Haiku | 79.8 | #571 |
| 11 | Llama 3.1 405B Instruct | 79.2 | #470 |
| 12 | Llama 3.3 70B Instruct | 77.2 | #541 |
| 13 | Gemini 2.0 Flash | 76.1 | #351 |
| 14 | GPT-4.1 Mini | 74.5 | #363 |
| 15 | GPT-4o Mini | 73.6 | #598 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tales-textworld-express · How It Works · Data refreshed daily, snapshot 2026-11-05.