AMAImedia/Qwen3.8-27B-Kimiko-2-NOESIS-BF16

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 30, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

AMAImedia/Qwen3.8-27B-Kimiko-2-NOESIS-BF16 is a 27 billion parameter merged language model based on the Qwen3.8 architecture, developed by AMAImedia as part of the NOESIS platform. This model is a Karcher Mean merge of several Qwen3.8 variants, demonstrating significant improvements in reasoning tasks like GSM8K (up to 96.4) and specialized gains in translation, intent classification, and multi-turn tool calling. It is designed for advanced agentic tasks, coding, and professional work, offering enhanced autonomous planning and environment feedback handling.

Loading preview...

Model Overview

AMAImedia/Qwen3.8-27B-Kimiko-2-NOESIS-BF16 is a 27 billion parameter language model developed by AMAImedia, founded by Ilia Bolotnikov, as part of the NOESIS Professional Multilingual Dubbing Automation Platform. This model is a merge of several Qwen3.8-based models, utilizing the Karcher Mean method to combine their strengths. It inherits the core capabilities of the Qwen3.8 series, including native vision-language understanding and flexible thinking control, with a context length of 32,768 tokens.

Key Differentiators and Performance

This merged model shows notable improvements in specific areas:

  • Reasoning: Achieves a GSM8K strict score of 96.4, a significant 29-point increase over its base Qwen3.8-27B. This indicates strong mathematical and logical reasoning capabilities.
  • Agentic and Coding Tasks: The base Qwen3.8-27B demonstrates strong performance in agentic coding (SWE-bench Pro: 61.7, DeepSWE 1.1: 42.2) and long-horizon office work (CoWorkBench: 70.7), which are foundational to this merge.
  • Specialized Gains: Internal benchmarks highlight substantial improvements in translation (+22.9), intent classification (+20.9), intent routing (+20.7), and multi-turn tool calling (+15.1).
  • Multilingual Support: Supports a wide array of languages, making it suitable for diverse linguistic applications.

Recommended Use Cases

This model is particularly well-suited for:

  • Complex Reasoning: Applications requiring advanced problem-solving, especially in mathematical and logical domains.
  • Agentic Workflows: Developing AI agents that need robust planning, tool use, and environment interaction.
  • Coding and Software Engineering: Tasks involving code generation, debugging, and repository-level understanding.
  • Multilingual Applications: Scenarios benefiting from its broad language support and enhanced translation capabilities.

Finetunes
1 models