Back to Models
ByteDance SeedOpen Source

Seed-OSS-36B

Seed-OSS-36B is a foundation model developed by ByteDance's Seed team. It is equipped with an approximate parameter scale of 360 billion and an extensive 512K context window.

Parameters

360.0B

Context Window

512K

License

Apache 2.0

Release Date

2025-08-20

API Pricing

API pricing for this model is not yet available

Strengths

  • Massive parameter scale of 360B
  • Very long context understanding up to 512K
  • Open license under Apache 2.0

Weaknesses

  • Large file size reaching 72GB
  • Requires high-performance computational resources
  • Potentially high inference costs

Use Cases

  • Ultra-long document analysis
  • Large-scale data context processing
  • Secondary development on open-source foundation

Deep Analysis

Architecture

Dense Transformer (36B)

Open-source from ByteDance Seed

Context Window

512K tokens

Native long context

Training Data

12T tokens

Release Date

August 21, 2025

License

Apache 2.0

GitHub Stars

875

ByteDance-Seed/seed-oss

Key Feature

Flexible thinking budget control

Dynamic reasoning length adjustment

Strengths

  • Native 512K long context
  • Flexible thinking budget control for efficiency
  • Apache 2.0 license
  • Strong reasoning, coding, and agent capabilities
  • Both Base and Instruct variants released
  • Research-friendly (with/without synthetic data variants)

Weaknesses

  • 36B parameter requires significant compute
  • Relatively new model
  • Limited production track record

Competitor Comparison

ModelArenaSWEGPQAPrice
Qwen 2.5 72B---Free (open)
Llama 3.1 70B---Free (open)
Mistral Large 3---Paid API

Seed-OSS-36B is ByteDance Seed Team's open-source language model with native 512K context and flexible thinking budget control. Released August 2025 under Apache 2.0, it achieves strong results on mainstream benchmarks while maintaining good practical performance at low cost.

Analysis generated: 2026-05-24