TIDE Trajectory - WebShop - Success: leaderboard
Trajectory-based diagnostics under the original unified agent interface, paper v2 Table 1. Five environments keep their own success rates and area under variation; AUV uses the printed 0-100 scale and measures progress across turns. Thinking and non-thinking backbones remain distinct. Separate from TIDE-Bench tool-integrated reasoning. This series reports Success Rate (%). Source: https://arxiv.org/html/2602.02196v2.
Metric: Success Rate (%). Source: arxiv.org. Saturation forecast: Around August 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 43.2 | #159 |
| 2 | DeepSeek V3.2 (Non-reasoning) | 41 | #210 (DeepSeek V3.2) |
| 3 | DeepSeek R1 | 38.4 | #262 |
| 4 | Gemini 2.5 Flash (Non-reasoning) | 35 | #255 (Gemini 2.5 Flash) |
| 5 | Llama 3.3 70B Instruct | 26.8 | #541 |
| 6 | GPT-OSS-120B | 26.2 | #346 |
| 7 | GLM-4 32B (0414) | 25.4 | #539 |
| 8 | Phi-4 (Reasoning) | 22.6 | #707 (Phi-4) |
| 9 | Llama 3.1 8B Instruct | 20.8 | #1017 |
| 10 | Phi-4 | 18.8 | #707 |
| 11 | GLM-4 9B Chat | 16.4 | #911 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tide-trajectory-webshop-success · How It Works · Data refreshed daily, snapshot 2026-11-05.