EMRB - Average: leaderboard
Average accuracy over EMRB's five levels: LLM code agents analyze raw I/Q electromagnetic captures by writing and running code, across 200 problems and 27 question types from signal detection to OFDM design (EMRB, arXiv 2608.24086v1, Appendix Table 5). Percent; higher is better; temperature 0 with provider default reasoning.
Metric: Accuracy (%; mean of five levels). Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 78.9 | #33 |
| 2 | Grok 4.5 | 77.8 | #54 |
| 3 | Claude Opus 4.7 | 70.3 | #51 |
| 4 | GLM-5.2 | 65.8 | #80 |
| 5 | Claude Sonnet 4.6 | 65.6 | #94 |
| 6 | Gemini 3.1 Pro (Preview) | 65.5 | #64 |
| 7 | DeepSeek V4 Pro | 62.6 | #112 |
| 8 | Gemini 3.5 Flash | 57.2 | #66 |
| 9 | MiniMax-M3 | 56.9 | #151 |
| 10 | MiMo-V2.5-Pro | 56.8 | #167 |
| 11 | DeepSeek V4 Flash | 56.4 | #155 |
| 12 | Kimi K2.6 | 50.6 | #113 |
| 13 | Llama 3.3 70B | 24.1 | #578 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=emrb-average · How It Works · Data refreshed daily, snapshot 2026-11-05.