roskosmos19/Dolphin-4B-Instruct-0409

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Dolphin-4B-Instruct-0409 by roskosmos19 is a 4 billion parameter instruction-tuned causal language model based on the Qwen3-4B architecture. It is specifically optimized for cost-efficiency and performance, featuring a reduced native context of 16k tokens and removed multimodal special tokens for a leaner footprint. This model is designed to deliver strong instruction following and high-quality answers at a minimal operational cost, making it ideal for applications requiring efficient inference.

Loading preview...

Dolphin-4B-Instruct-0409: Optimized Price/Performance 4B Model

Dolphin-4B-Instruct-0409 is an instruction-tuned language model developed by roskosmos19, built upon the robust Qwen3-4B architecture. Its primary focus is to deliver the best possible capability at the lowest operational cost, making it a highly efficient choice for various applications.

Key Optimizations and Features

This model distinguishes itself from the original Qwen3-4B-Instruct through several strategic optimizations:

  • Optimized Context Length: While the original Qwen3-4B offered 262k context, Dolphin-4B-Instruct-0409 is tuned for a 16k native context, which covers over 95% of typical real-world use cases while significantly reducing KV-cache VRAM consumption.
  • Leaner Architecture: Multimodal special tokens (e.g., for vision) have been removed to create a more streamlined and efficient model.
  • Enhanced Instruction Following: It features a stronger system prompt and generation defaults specifically tuned for higher quality, more precise, and less repetitive answers.
  • Efficiency: Despite maintaining the same 4 billion parameters and intelligence as its base, these optimizations result in a model that is significantly cheaper and faster to run.

Technical Specifications

  • Architecture: Qwen3ForCausalLM
  • Parameters: 4.0 Billion
  • Context Window: 16,384 tokens
  • Recommended Data Types: bfloat16 or float16
  • Recommended Quantization: Q4_K_M / AWQ / GPTQ for optimal speed and quality.

Ideal Use Cases

This model is particularly well-suited for developers and applications where:

  • Cost-efficiency is paramount without sacrificing instruction-following capabilities.
  • Efficient inference on limited hardware resources is required.
  • High-quality, focused text generation is needed for tasks like chatbots, content generation, or summarization within a typical context window.

Finetunes
1 models