TeleCom-Bench - Tool Invocation: leaderboard
Knowledge-application task of TeleCom-Bench (Xiao et al., ZTE Corporation, KDD 2026) mined from live network-operation agent trajectories: the model writes the sequence of tool calls, with parameters, that carries out a network-operation workflow. Exact match in percent, with reasoning enabled, temperature 0.7 and a majority vote over three samples; higher is better.
Metric: Exact match (%; tool-invocation workflow generation from a fixed tool library). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 (Thinking) | 94.06 | #403 (DeepSeek V3.2) |
| 2 | Qwen 3 32B (Thinking) | 84.71 | #960 (Qwen 3 32B) |
| 3 | GLM-4.7 (Thinking) | 56.5 | #365 (GLM-4.7) |
| 4 | Qwen 3 235B A22B (Thinking) | 45.54 | #622 (Qwen 3 235B A22B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=telecom-bench-tool-invocation · How It Works · Data refreshed daily, snapshot 2026-11-12.