seed-oss-36b-instruct.
ByteDance Seed Aug 2025 Open weights Mid-size
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.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| RULER | 94.6% | #10 of 42 | Lab model card | |
| MMLU | 87.4% | #37 of 131 | Lab model card | |
| MMLU-Pro | 82.7% | #52 of 167 | Lab model card | |
| IFEval | 85.8% | #76 of 135 | Lab model card | |
| MMMLU | 78.4% | #48 of 71 | Lab model card | |
| SuperGPQA | 55.7% | #24 of 35 | Lab model card | |
| SimpleQA | 9.70% | #51 of 60 | Lab model card | |
| Humanity's Last Exam | 10.1% | #141 of 154 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AIME 2024 | 91.7% | #11 of 88 | Lab model card | |
| GPQA Diamond | 71.5% | #116 of 229 | Epoch AI run | |
| AIME 2025 | 84.7% | #75 of 143 | Lab model card | |
| OTIS Mock AIME 2024-2025 | 67.5% | #115 of 212 | Epoch AI run | |
| Beyond AIME | 65% | #8 of 14 | Lab model card | |
| KOR-Bench | 70.6% | #3 of 4 | Lab model card | |
| ARC-AGI-2 | 1.45% | #24 of 24 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench v6 | 67.4% | #63 of 90 | Lab model card | |
| SWE-bench Verified | 56% | #119 of 155 | Lab model card | |
| Multi-SWE-bench | 17% | #10 of 10 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau-bench | 58.2% | #14 of 23 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveSecBench | 30.3 | #26 of 34 | Official leaderboard | |
| AIR-Bench 2024 | 75.6% | #2 of 2 | Lab model card |
The Self-Reported Index fits one score to the 17 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 132.0 to 136.7. How we compute it.
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.
The benchmarks the lab chose to headline, with the scores it gave.