21.3K Downloads

seed-oss-36b

bytedance/

seed-oss-36b

36B

seed_oss

Advanced reasoning model from ByteDance with flexible "thinking budget" control and ability to reflect on the length of its own reasoning

Trained for Ttool use

Reasoning

Last Updated28 days ago

Requires min. Min.21GB

README

Seed-OSS 36B

Advanced reasoning model with flexible thinking budget control and native 512K context support

36B parameters with GQA attention architecture, designed for powerful long-context reasoning, agentic tasks, and general capabilities

Features dynamic reasoning length control, allowing users to adjust thinking budget from 512 tokens to unlimited based on task complexity

Excels at mathematical reasoning, coding tasks, tool usage, and agentic workflows including SWE-Bench and issue resolution

Achieves state-of-the-art performance on multiple benchmarks including MATH (81.7%), LiveCodeBench (67.4%), and RULER long-context (94.6%)

Optimized for international use cases with Apache 2.0 license and research-friendly design

custom fields

Special features defined by the model author

Thinking Budget

: select

(default=-1)

Sets the maximum number of tokens the model can use for internal reasoning

sources

The underlying model files this model uses

Based on

GGUF

MLX

MLX

MLX

MLX

When you download this model, LM Studio picks the source that will best suit your machine (you can override this)

config

Custom configuration options included with this model

Reasoning Section Parsing

{ "enabled": true, "startString": "<seed:think>", "endString": "</seed:think>" }

Temperature

1.1

Top P Sampling

0.95