NetAgentBench: leaderboard
NetAgentBench (2026) has LLM agents configure emulated network devices over up to 100 turns on CCNA-, CCNP OSPF- and CCIE BGP-level tasks, 25 runs each, and checks the resulting device state rather than the commands. This board is the mean combined score from 0 to 1, the equal-weight mean of property completeness, re-application robustness and command validity; higher is better.
Metric: Mean combined score (0 to 1): the equal-weight mean of property completeness of the final device state, re-application robustness and the share of valid commands, averaged over 75 runs per model (three network-configuration tasks at CCNA, CCNP OSPF and CCIE BGP level, 25 repetitions each) on emulated routers, at most 100 turns and 30 minutes per run; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 0.508 | #178 |
| 2 | Qwen 3 Coder 30B A3B Instruct | 0.31 | #1012 |
| 3 | Llama 3.3 70B Instruct | 0.231 | #1385 |
| 4 | Qwen3 Coder Next | 0.221 | #725 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=netagentbench · How It Works · Data refreshed daily, snapshot 2026-11-21.