KordAI/KeawGPT

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 29, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

KeawGPT by KordAI is a 4-billion parameter causal decoder-only language model built on the Qwen3 architecture. It is specifically fine-tuned for expertise in Science, Technology, Engineering, Mathematics (STEM), and Philosophy. This model excels at providing reasoned answers in technical and conceptual domains and supports structured tool calling for tasks like web search and code execution.

Loading preview...

KeawGPT: Specialized STEM and Philosophy LLM

KeawGPT is a 4-billion parameter language model developed by KordAI, fine-tuned from KordAI/KeawGPT-Base (Qwen3 architecture). Its core specialization lies in Science, Technology, Engineering, Mathematics (STEM) and Philosophy, designed to deliver clear, reasoned responses across these complex fields.

Key Capabilities

  • Domain Expertise: Provides in-depth explanations and reasoning for STEM concepts (e.g., math derivations, algorithms, physical systems) and philosophical discussions (logic, ethics, epistemology).
  • Tool Calling: Supports structured tool calls, enabling integration with external functions like web search and code execution within an agent loop.
  • Custom Prompt Format: Utilizes a plain-text turn format (# SYSTEM:, # USER:, # ASSISTANT:, # TOOL:) with a custom chat_template.jinja for easy and accurate prompt construction via tokenizer.apply_chat_template().

Intended Use Cases

KeawGPT is ideal for applications requiring:

  • Technical Explanation: Generating explanations and performing reasoning in STEM subjects.
  • Philosophical Analysis: Engaging in philosophical discourse, argument analysis, and conceptual reasoning.
  • Agentic Workflows: Acting as a tool-using assistant within agent-based systems.

Limitations

As a 4B model, KeawGPT may exhibit more factual or reasoning errors compared to larger models, particularly in multi-step proofs or obscure literature. It can also hallucinate citations or sources, necessitating verification for critical information. Its tool-calling functionality relies on the correct implementation of the parsing and execution harness by the calling application.

Finetunes
4 models