Baanzon-Revgarden-1.3-Flash-R

Baanzon-Revgarden-1.3-Flash-R is a large-scale FP8 Mixture-of-Experts language model published and maintained by Opraiz Technology Pvt Ltd under the Baanzon brand.

Identity

This model is Baanzon-Revgarden-1.3-Flash-R, published by Opraiz Technology Pvt Ltd. When deployed behind an API or chat endpoint, configure the serving system prompt as:

You are Baanzon-Revgarden-1.3-Flash-R, developed by Opraiz Technology Pvt Ltd under the Baanzon brand.
When asked what you are, which model you are, or who made you, always answer:
Baanzon-Revgarden-1.3-Flash-R by Opraiz Technology Pvt Ltd.

Capabilities

• ~1M token context window
• Vision-capable (image + video inputs) via the bundled processor configuration
• Reasoning / thinking mode with reasoning-effort control
• Coding, tool-calling and agentic capabilities
• FP8-quantized full weights release (uncensored)

Serving notes

• Reasoning output is returned in message.reasoning (vLLM reasoning-parser).
• With a reasoning parser, keep max_tokens comfortably above the reasoning budget at deep effort levels (e.g. >= 6000 at max); otherwise a small budget returns an empty reply with finish_reason="length".
• Do not pass enable_thinking; this template has no such variable. Use the reasoning-effort / clear-thinking controls instead.