TALES - ALFWorld: leaderboard
TALES ALFWorld subset: text-only household tasks from ALFWorld, where the only terminal reward is task completion, invalid actions return just 'Nothing happened', and the agent can hold one object at a time. 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 (Medium) | 88.3 | #132 (O3) |
| 2 | GPT-4.1 | 83.3 | #258 |
| 3 | Claude 3.7 Sonnet (Thinking) | 83.3 | #259 (Claude 3.7 Sonnet) |
| 4 | Claude 3.7 Sonnet | 81.7 | #259 |
| 5 | O3 (High) | 81.7 | #132 (O3) |
| 6 | O3 (Low) | 70 | #132 (O3) |
| 7 | GPT-4o | 56.7 | #357 |
| 8 | Qwen 3 32B | 48.3 | #449 |
| 9 | Qwen 2.5 72B Instruct | 36.7 | #469 |
| 10 | Llama 3.1 405B Instruct | 31.7 | #470 |
| 11 | O1 | 28.3 | #190 |
| 12 | Claude 3.5 Haiku | 26.7 | #571 |
| 13 | Gemini 2.0 Flash | 20 | #351 |
| 14 | Llama 3.3 70B Instruct | 15 | #541 |
| 15 | Llama 3.1 70B Instruct | 8.3 | #567 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tales-alfworld · How It Works · Data refreshed daily, snapshot 2026-11-05.