6G-Bench - Intent and Policy Reasoning: leaderboard
Metric: Accuracy (%) on the intent and policy reasoning tasks (T1, T2, T3, T12, T15) of 6G-Bench's 3,722 expert-validated four-option multiple-choice questions on network-level semantic reasoning for AI-native 6G networks, deterministic single-shot answers (temperature 0, one letter in a JSON object), group score is the unweighted mean of its task accuracies; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3 Coder Next | 88.6 | #321 |
| 2 | Llama 4 Maverick | 88.1 | #451 |
| 3 | GPT-5.2 Instant | 87.8 | #205 |
| 4 | DeepSeek V3.2 | 87.6 | #198 |
| 5 | Ministral 3 14B | 87 | #636 |
| 6 | GPT-5.2 Codex | 86.8 | #89 |
| 7 | DeepSeek V3.2 Exp | 86.3 | #227 |
| 8 | Olmo 3.1 32B Instruct | 86.1 | #754 |
| 9 | Hermes 4 70B | 86 | #489 |
| 10 | Claude Haiku 4.5 | 85.3 | #271 |
| 11 | Ministral 3 8B | 85.3 | #676 |
| 12 | Qwen 3 VL 32B Instruct | 85.2 | #276 |
| 13 | Phi-4 | 85.1 | #701 |
| 14 | Qwen 3 235B A22B 2507 Instruct | 84.5 | #291 |
| 15 | GPT-4o Mini | 84.4 | #588 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=6g-bench-intent-and-policy-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-11.