Seed-OSS-36B-Instruct: benchmark results
ByteDance's open Seed-OSS 36B instruct model (512K context). Provider: ByteDance. Released 2025-08-20. Access: Open.
Unified ELO 1615 ± 15, rank #386 of 1639 rated models, from 203 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ReLE - Education - Middle School Subjects | 62.5 | Accuracy (%) | 100 |
| SuiChat-CN - Few-Shot | 77.3 | Macro F1 (%) of suicide-risk level classification of Chinese | 100 |
| OpenCompass Reasoning - Puzzle - Chinese (CompassBench 2510) | 93.1 | Score (%) | 98.1 |
| SuiChat-CN | 74.5 | Standard F1 (self-reported) | 90.7 |
| SuiChat-CN - Without Keywords | 74.2 | Macro F1 (%) of suicide-risk level classification of Chinese | 88.4 |
| AA LiveCodeBench | 76.51 | Pass@1 (%) | 87.4 |
| MolecularIQ - Indexing | 25.6 | Mean Accuracy (%) | 86.5 |
| SuiChat-CN - Without Context | 71.3 | Macro F1 (%) of suicide-risk level classification of Chinese | 86 |
| OpenCompass LLM - Math (CompassBench 2510) | 64.5 | Score (%) | 84.6 |
| OpenCompass Math - Competition - English (CompassBench 2510) | 75 | Score (%) | 84.6 |
| AA AIME 2025 | 84.67 | Accuracy (%) | 81.9 |
| MolecularIQ - Counting | 26.8 | Mean Accuracy (%) | 81.1 |
Interactive version: theaggregate.ai/model?slug=seed-oss-36b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-09.