Flexan/Blake-Haiku-1

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 21, 2026License:cc-by-sa-4.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Flexan/Blake-Haiku-1 is a 0.6 billion parameter instruct large language model, LoRA fine-tuned from Qwen/Qwen3-0.6B. It is specifically trained to generate human-like conversational text, mimicking traditional chat platforms. This model was developed primarily as a test of a new runtime environment for training larger models on personal hardware, rather than for production deployment. It does not support reasoning or tool-calling capabilities.

Loading preview...

Overview

Blake Haiku 1 is a 0.6 billion parameter instruct large language model, fine-tuned from Qwen/Qwen3-0.6B. Its primary design goal was to enable human-like conversational interactions, mimicking the style of traditional chat platforms like Discord, where users can send multiple messages before receiving a response. This model was developed by Flexan as a technical exercise to test a new runtime environment for training larger models on personal hardware, specifically on Windows 11 CUDA.

Key Characteristics

  • Conversational Style: Trained to produce responses that resemble human chat conversations, with each line potentially representing a new message.
  • Base Model: Utilizes Qwen/Qwen3-0.6B as its foundational architecture.
  • Limited Scope: Explicitly trained without support for reasoning or tool-calling, and its internal <think>...</think> tags are always empty.
  • Chat Format: Employs the ChatML format for interactions, with a recommended system prompt for a specific persona (e.g., "You're Moke, a user chatting with random people on Discord.").

Intended Use and Limitations

Blake Haiku 1 is not intended for production deployment due to its minimal training data and its primary purpose as a technical testbed. The developer explicitly states it was archived for this reason and is not meant for general use. Users should be aware of its small dataset and lack of advanced reasoning capabilities. There are no plans for a "Blake Haiku 2".

Finetunes
2 models