Seed 2.0 Lite: benchmark results
ByteDance's lightweight Seed 2.0 tier. Provider: ByteDance. Released 2026-02-14. Access: API.
Aggregate Rating 1731 ± 8, rank #145 of 1656 rated models, from 273 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Beyond Function Calling | 51.3 | Overall (self-reported) | 100 |
| EmoTrans - Emotion State Identification | 41.2 | Accuracy (%) on Emotion State Identification (724 questions: | 100 |
| EmoTrans - Next Emotion Prediction | 36.2 | Accuracy (%) on Next Emotion Prediction (826 questions: the | 100 |
| ODU-Bench (Audio-Only) | 77.3 | Weighted demand-understanding score (%; 0.15 demand detectio | 100 |
| ODU-Bench (Audio-Only) - Demand Localization | 71.4 | Temporal IoU (%; overlap of the predicted and reference dema | 100 |
| ODU-Bench (Audio-Only) - Key-Point Coverage | 78 | Key-point hit rate (%; share of reference key points on stru | 100 |
| ODU-Bench (Audio-Only) - Transcription | 93.2 | Transcript accuracy (%; max(0, 1 - pooled CER or WER) of the | 100 |
| ODU-Bench (Audio-Visual) - Demand Localization | 82.8 | Temporal IoU (%; overlap of the predicted and reference dema | 100 |
| Qwen3.8 Omni Report - ProactiveVideoQA | 68.3 | Score (%) | 100 |
| Qwen3.8 Omni Report - Vox-Infinity - Personal Monologues | 69.3 | Personal Monologues (%) | 100 |
| ReLE - Education - Middle School Subjects | 59.5 | Accuracy (%) | 98.4 |
| ReLE - Reasoning and Mathematics | 85.8 | Accuracy (%) | 96.8 |
Interactive version: theaggregate.ai/model?slug=seed-2-0-lite · How It Works · Data refreshed daily, snapshot 2026-10-20.