- GPQA Diamond 88.4
- SuperGPQA 69.6
- HLE (no tool, text only) 25.7
- BeyondAIME 79
- FrontierSci-olympiad 72
- Superchem (text-only) 55
- BABE 57.9
- CL-Bench 20.1
- MultiChallenge 69.9
- WideSearch 70.3
- BrowseComp 64
- ResearchRubrics 59.2
- XPert Bench 56.8
- SkillsBench 43.7
- GDPval 53.1
- FinSearchComp 63.8
- SWE Multilingual 66.6
- SWE-Bench Pro 46.6
- NL2Repo-Bench 28.7
- PaperBench 52.5
- Terminal Bench 2.0 43.3
- MathVision 89.8
- MMMU_Pro 78.4
- HiPhO 83.8
- MedXpertQA-MM 79.6
- BabyVision 64.7
- VLMBias 80.6
- SimpleVQA 72.7
- WorldVQA 50.2
- CharXiv-RQ 82.4 (CharXiv-DQ 94.5) 82.4
- ERQA 71.5
- OSWorld-Verfied 64.4
- MobileWorld 64.6
- VideoMMMU 88.3
- MMVU 76.7
- VideoSimpleQA (VideoSimpleQA-v2 69.0) 71.7
- VideoHolmes 67.4
- Minerva 68.5
- TVBench 80.4
- TOMATO 72.5
- EgoTempo 68.4
- MotionBench 72.4
- VideoMME 89
- VideoMMEv2 (subtitle setting not stated) 64.9
- LongVideoBench 79
- LVBench 76.4
- OVBench 63.2
- OVOBench 75.4
- OmniVideoBench 61.7
- JointAVBench 69.5
- WorldSense 67.3
- MMSU 86.5
- WildSpeech 75.8
- Mean of WenetSpeech test-net 4.47 / test-meeting 5.31 (CER, lower is better) 4.89
- Mean of Librispeech test-clean 1.07 / test-other 2.17 (WER, lower is better) 1.62
- Fleurs (15 langs) (zh/en<->xx) speech translation
- VideoReasonBench 59.4
- ContPhy 62.4
- Morese-500 34.6
- CGBench 65.5
- VideoEval-Pro 49.5
- ODVBench 66
- LiveSports-3K 78.1
- ViSpeak 87
- CrossVid 63.7
- AVMeme 69.5
Seed2.0 Lite (0428)
Seed2.0 Lite (0428) ranks first on 2 of the 45 benchmarks with five or more models scored.
63
Benchmarks with a score
2
Ranked first, out of five or more models
15
In the top three
ByteDance Seed Apr 2026 Seed2.0 Lite (0428)
Closed weights Mid-size
148.8
#103 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- MMMU-Pro 78.4% #25 /90
- NL2Repo-Bench 28.7% #28 /30
- SimpleVQA 72.7% #7 /30
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MultiChallenge | 69.9% | #4 of 42 | Lab model card | |
| SuperGPQA | 69.6% | #7 of 35 | Lab model card | |
| SimpleVQA | 72.7% | #7 of 30 | Lab model card | |
| WorldSense | 67.3% | #1 of 4 | Lab model card | |
| Humanity's Last Exam | 25.7% | #91 of 154 | Lab model card | |
| WideSearch | 70.3% | #12 of 19 | Lab model card | |
| Video SimpleQA | 71.7% | #3 of 4 | Lab model card | |
| CL-bench | 20.1% | #5 of 6 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 88.4% | #49 of 302 | Lab model card | |
| Beyond AIME | 79% | #3 of 14 | Lab model card | |
| FrontierScience | 72% | #5 of 8 | Lab model card | |
| HiPhO | 83.8% | #3 of 4 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PaperBench | 52.5% | #5 of 15 | Lab model card | |
| SWE-bench Multilingual | 66.6% | #52 of 73 | Lab model card | |
| SWE-Bench Pro | 46.6% | #74 of 81 | Lab model card | |
| NL2Repo-Bench | 28.7% | #28 of 30 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BrowseComp | 64% | #53 of 81 | Lab model card | |
| OSWorld-Verified | 64.4% | #27 of 41 | Lab model card | |
| Terminal-Bench 2.0 | 43.3% | #48 of 63 | Lab model card | |
| ResearchRubrics | 59.2% | #5 of 6 | Lab model card | |
| SkillsBench | 43.7% | #13 of 14 | Lab model card | |
| MobileWorld | 64.6% | #3 of 3 | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedXpertQA-MM | 79.6% | #2 of 21 | Lab model card | |
| FinSearchComp | 63.8% | #4 of 7 | Lab model card | |
| XpertBench | 56.8% | #2 of 3 | Lab model card | |
| GDPval | 53.1% | #9 of 10 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Video-MMMU | 88.3% | #2 of 36 | Lab model card | |
| WorldSense (omnimodal) | 67.3% | #1 of 12 | Official leaderboard | |
| Video-MME | 89% | #5 of 51 | Lab model card | |
| ERQA | 71.5% | #6 of 36 | Lab model card | |
| MMSU | 86.5% | #1 of 6 | Lab model card | |
| MATH-Vision (MathVision) | 89.8% | #10 of 51 | Lab model card | |
| MMVU | 76.7% | #6 of 26 | Lab model card | |
| LiveSports-3K | 78.1% | #1 of 4 | Lab model card | |
| MMMU-Pro | 78.4% | #25 of 90 | Lab model card | |
| CharXiv | 82.4% | #23 of 75 | Lab model card | |
| LVBench | 76.4% | #10 of 31 | Lab model card | |
| MotionBench | 72.4% | #3 of 9 | Lab model card | |
| MINERVA | 68.5% | #2 of 6 | Lab model card | |
| CG-Bench | 65.5% | #1 of 3 | Lab model card | |
| ViSpeak-Bench | 87% | #1 of 3 | Lab model card | |
| LongVideoBench | 79% | #4 of 11 | Lab model card | |
| EgoTempo | 68.4% | #2 of 5 | Lab model card | |
| TOMATO | 72.5% | #2 of 5 | Lab model card | |
| TVBench | 80.4% | #3 of 7 | Lab model card | |
| Video-Holmes | 67.4% | #3 of 6 | Lab model card | |
| VideoMME v2 (with subtitles) | 64.9% | #3 of 6 | Lab model card | |
| VideoEval-Pro | 49.5% | #1 of 2 | Lab model card | |
| BabyVision | 64.7% | #14 of 26 | Lab model card | |
| OVO-Bench | 75.4% | #3 of 5 | Lab model card | |
| VLMsAreBiased | 80.6% | #3 of 5 | Lab model card | |
| WorldVQA | 50.2% | #6 of 9 | Lab model card | |
| CrossVid | 63.7% | #2 of 3 | Lab model card | |
| VideoReasonBench | 59.4% | #2 of 3 | Lab model card | |
| OVBench | 63.2% | #4 of 5 | Lab model card | |
| ContPhy | 62.4% | #3 of 3 | Lab model card | |
| JointAVBench | 69.5% | #2 of 2 | Lab model card | |
| MORSE-500 | 34.6% | #2 of 2 | Lab model card | |
| ODV-Bench | 66% | #2 of 2 | Lab model card | |
| OmniVideoBench | 61.7% | #2 of 2 | Lab model card | |
| AVMeme Exam | 69.5% | #1 of 1 | Lab model card |
Its lab's numbers put it at 148.8, number 103 of 404
The Self-Reported Index fits one score to the 29 capability results ByteDance Seed reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 147.1 to 150.5. How we compute it.
- MATH-Vision (MathVision) 89.8
- Video-MME 89
- GPQA Diamond 88.4
- Video-MMMU 88.3
- CharXiv 82.4
- MedXpertQA-MM 79.6
- Beyond AIME 79
- LongVideoBench 79
- MMMU-Pro 78.4
- MMVU 76.7
- LVBench 76.4
- SimpleVQA 72.7
- MotionBench 72.4
- FrontierScience 72
- WideSearch 70.3
- MultiChallenge 69.9
- SuperGPQA 69.6
- SWE-bench Multilingual 66.6
- BabyVision 64.7
- OSWorld-Verified 64.4
- BrowseComp 64
- GDPval 53.1
- PaperBench 52.5
- WorldVQA 50.2
- SWE-Bench Pro 46.6
- SkillsBench 43.7
- Terminal-Bench 2.0 43.3
- NL2Repo-Bench 28.7
- Humanity's Last Exam 25.7
Its lab's number next to a leaderboard
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| MATH-Vision (MathVision) | 89.8% MathVision | 89.8% Leaderboard | Same |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.