Azure99/Blossom-V6.4-35B-A3B

TEXT GENERATIONConcurrency Cost:3Model Size:35.1BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 9, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Blossom-V6.4-35B-A3B by Azure99 is a 35.1 billion parameter open-source conversational large language model. It is designed for general-purpose use, providing reproducible post-training data and maintaining multimodal capabilities. This model is optimized for cost-effective local accessibility and robust conversational applications.

Loading preview...

Azure99/Blossom-V6.4-35B-A3B Overview

Blossom-V6.4-35B-A3B is a 35.1 billion parameter open-source conversational large language model developed by Azure99. It is part of the Blossom-V6.4 series, which largely follows the V6.3 training recipe, incorporating a small number of multimodal samples to preserve its multimodal capabilities. The model emphasizes reproducible post-training data and aims to provide a powerful, cost-effective, and locally accessible general-purpose solution.

Key Capabilities & Training

  • Conversational AI: Designed as a robust conversational large language model.
  • Multimodal Preservation: Includes multimodal samples to retain capabilities from base models.
  • Reproducible Data: Focuses on providing reproducible post-training data.
  • Advanced Data Synthesis: Employs a sophisticated data synthesis workflow using multiple models (Deepseek-V3.1, Gemini 2.5 Flash, Qwen3-235B-A22B-Instruct-2507) for response generation and verification. This includes cross-evaluation for subjective scenarios and correctness checks for objective tasks like mathematics.
  • Rule-Based Filtering: Utilizes N-Gram filtering to remove repetitive data and discards toxic content.

Use Cases

  • General-Purpose Conversational Applications: Suitable for a wide range of dialogue-based tasks.
  • Local Deployments: Optimized for cost-effective and locally accessible implementations.
  • Research and Development: Provides a foundation for further research with its reproducible data and open-source nature.