TIDE-Bench - Interaction: leaderboard

The Interaction track of TIDE-Bench (2026): the model acts as a merchant service agent with five user-defined e-commerce tools for up to ten turns with a simulated customer in 100 scripted scenarios, and a Kimi K2 judge scores task success, customer satisfaction and tool-call behaviour from 0 to 5. Mean of the three scores (0-5); higher is better.

Metric: Average customer score (out of 5): mean of the customer-facing success, satisfaction and tool-call behaviour scores (each 0-5) a kimi-k2-0905-preview judge gives the full trajectory of a merchant service agent over up to ten turns with five user-defined e-commerce tools and an API-model customer, 100 scripted scenarios in five service categories (TIDE-Bench Interaction track); higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 32B4.53#972
2Qwen 3 8B4.08#1873
3Qwen 3 4B3.79#2348
4GPT-5.43.57#135
5GPT-52.77#178
6Tool-Star2.46
7SearchR12.1
8GPT-4o1.62#751

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=tide-bench-interaction · How It Works · Data refreshed daily, snapshot 2026-11-21.