Seed-OSS-36B drops with 512K context
ByteDance’s Seed Team released Seed-OSS-36B and Seed-OSS-36B-Instruct as Apache-2.0 open models with native 512K context, controllable thinking budgets, and strong reasoning, coding, and agent benchmarks.
Seed-OSS-36B stands out as an open-model release aimed at builders who want long context, tunable reasoning cost, and self-hosted flexibility rather than another generic chat wrapper.
- –The 512K native context window makes the model more practical for workflows that need large document sets and long-running prompts
- –Thinking-budget control is a useful product feature because it gives developers a direct lever over reasoning depth, latency, and inference cost
- –ByteDance positions the model as agent-capable, with benchmarks covering tool use and software-task settings alongside reasoning and coding
- –Apache-2.0 licensing makes fine-tuning and self-hosted deployment much easier for teams that do not want to start on a closed API
- –At 36B parameters, deployment is still substantial enough that infrastructure planning matters before production use
DISCOVERED
83d ago
2026-03-06
PUBLISHED
83d ago
2026-03-06
RELEVANCE
AUTHOR
Discover AI
