benchmarks
Seed-OSS-36B-Instruct is ByteDance Seed's open-weight, 36-billion-parameter dense language model, released under the Apache 2.0 license. It supports a documented 512K-token context window and configurable thinking budgets for text-based reasoning, coding and agent applications.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| GPQA DiamondPhD-level science questions | 71.5%30 Aug 2026 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 67.5%28 Aug 2026 | View |
Weighted mean of 2 verified benchmark results: GPQA Diamond 71.5%, OTIS Mock AIME 2024-25 67.5%.
Capabilities
- Processes text within a documented 512K-token context window.
- Supports configurable thinking budgets to balance reasoning effort against generation cost.
- Supports text-based reasoning and mathematical problem solving.
- Supports code generation and coding assistance.
- Supports tool-use and agent workflows through external integrations.
- Provides open weights under Apache 2.0 for self-hosting and customization.
What improved
- Not yet confirmed by public reporting.
Known flaws & limitations
- Native image, audio and video understanding are not documented capabilities.
- Self-hosting requires infrastructure sized for a 36-billion-parameter dense model.
- Long-context inference adds memory and processing requirements beyond storing the model weights.
- Extended thinking consumes additional generated tokens and inference resources.
Risks, safety & security issues
- Not yet confirmed by public reporting.
Good fit for
- Self-hosted text assistants where deployment control and open licensing matter.
- Long-document analysis and text synthesis within the documented context limit.
- Coding and mathematical reasoning workflows with independently checked outputs.
- Tool-integrated assistants with configurable reasoning budgets.
Poor fit for
- Applications requiring native image, audio or video understanding.
- On-device deployments with insufficient memory for a 36-billion-parameter model.
- Strict low-latency applications that require extensive thinking on every request.
References
Profile last reviewed 10/16/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.