TeleChat2: benchmark results
China Telecom AI Research Institute's TeleChat2-115B, an open 115B Chinese-English model trained on 10 trillion tokens, using domestically-trained (Ascend) compute. Provider: China Telecom. Released 2024-09-20. Access: Open.
Aggregate Rating 1558 ± 43, rank #580 of 1635 rated models, from 57 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SuperCLUE-Agent - Long & Short-term Memory | 81.11 | Score | 90 |
| OpenCompass Language - Creation - Chinese (CompassBench 2411) | 59.2 | Score (%) | 87.9 |
| OpenCompass Reasoning - Common Sense - Chinese (CompassBench 2411) | 67 | Score (%) | 87.9 |
| OpenCompass LLM - Instruction Following - Chinese (CompassBench 2411) | 57.9 | Score (%) | 84.8 |
| OpenCompass Math - College - Chinese (CompassBench 2411) | 42.2 | Score (%) | 84.8 |
| OpenCompass Reasoning - Common Sense - English (CompassBench 2411) | 65.2 | Score (%) | 84.8 |
| OpenCompass Reasoning - Social - Chinese (CompassBench 2411) | 59.4 | Score (%) | 84.8 |
| OpenCompass Math - College - English (CompassBench 2411) | 44.6 | Score (%) | 83.3 |
| OpenCompass LLM - Language (CompassBench 2411) | 58.9 | Score (%) | 81.8 |
| OpenCompass LLM - Language - Chinese (CompassBench 2411) | 60.5 | Score (%) | 81.8 |
| OpenCompass LLM - Language - English (CompassBench 2411) | 57.3 | Score (%) | 81.8 |
| OpenCompass Language - Dialogue - English (CompassBench 2411) | 63.5 | Score (%) | 81.8 |
Interactive version: theaggregate.ai/model?slug=telechat2 · How It Works · Data refreshed daily, snapshot 2026-10-15.