Step 2 16K: benchmark results
StepFun's Step-2-16K, an early preview of their trillion-parameter mixture-of-experts model, accessible through an OpenAI-compatible API with a 16K context window. Provider: StepFun. Released 2024-07-04. Access: API.
Aggregate Rating 1603 ± 24, rank #422 of 1635 rated models, from 186 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenCompass Agent - T-Eval - English (CompassBench 2409) | 88.3 | Score (%) | 100 |
| OpenCompass Knowledge - Engineering - Chinese (CompassBench 2504) | 95 | Score (%) | 100 |
| OpenCompass Language - Creation - Chinese (CompassBench 2409) | 56.3 | Score (%) | 100 |
| OpenCompass Language - NLP - Chinese (CompassBench 2409) | 61.6 | Score (%) | 100 |
| OpenCompass Knowledge - Engineering - English (CompassBench 2411) | 86.7 | Score (%) | 98.5 |
| OpenCompass Reasoning - Common Sense - Chinese (CompassBench 2409) | 58.2 | Score (%) | 98.3 |
| OpenCompass Knowledge - Science - English (CompassBench 2411) | 84.9 | Score (%) | 97 |
| OpenCompass LLM - Knowledge - Chinese (CompassBench 2411) | 72.3 | Score (%) | 97 |
| OpenCompass Code - Code QA - English (CompassBench 2409) | 64.5 | Score (%) | 96.6 |
| OpenCompass LLM - Instruction Following (CompassBench 2409) | 64.8 | Score (%) | 96.6 |
| OpenCompass Language - Dialogue - English (CompassBench 2409) | 58.5 | Score (%) | 96.6 |
| OpenCompass Language - NLP - English (CompassBench 2409) | 57.8 | Score (%) | 96.6 |
Interactive version: theaggregate.ai/model?slug=step-2-16k · How It Works · Data refreshed daily, snapshot 2026-10-15.