ManniX-ITA/Qwen3.8-27B-Omnimerge-v6

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 3, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

ManniX-ITA/Qwen3.8-27B-Omnimerge-v6 is a 27 billion parameter language model, a task-arithmetic merge of three Qwen3.6 fine-tunes onto the Qwen3.8-27B base, utilizing an omnimerge_v2 method. This model maintains a 32768 token context length and incorporates the Qwen3.8 vision tower and MTP head. It demonstrates a significant improvement in code generation, specifically on the LiveCodeBench benchmark, and enhanced tool-calling safety compared to its predecessor, Qwen3.6-27B-Omnimerge-v4.

Loading preview...

Overview

ManniX-ITA/Qwen3.8-27B-Omnimerge-v6 is a 27 billion parameter model created by ManniX-ITA, built upon the Qwen3.8-27B base. It is a task-arithmetic merge, combining three Qwen3.6 fine-tunes using the omnimerge_v2 method. This approach applies deltas computed from Qwen3.6 sources to the newer Qwen3.8 base, ensuring the integration of generational upgrades without backward degradation. The model retains the Qwen3.8 base's tokenizer, chat template, MTP head, and vision tower, with the vision projector being bit-identical to the Qwen3.8 base.

Key Capabilities

  • Enhanced Code Generation: Shows a notable improvement of +6.5 percentage points on the LiveCodeBench (v6, 77q) benchmark compared to Omnimerge-v4.
  • Superior Tool-Calling Safety: Achieves a leading score in tool-eval-bench hardmode, with significantly fewer safety-critical failures (3 vs 9 for the base and 16 for v4). It specifically excels in Safety & Boundaries (89.2%).
  • Vision-Capable: Inherits the vision tower from the Qwen3.8 base, supporting multimodal applications.
  • Robust Merging: Utilizes a verified merging process that correctly transplants fine-tuned deltas from Qwen3.6 sources to the Qwen3.8 base, preserving vocabulary and vision components.

Good for

  • Developers requiring a 27B parameter model with strong code generation capabilities.
  • Applications demanding high safety and reliability in tool-calling scenarios.
  • Use cases benefiting from a multimodal model with integrated vision features.
  • Users seeking a model that builds upon the latest Qwen3.8 base while incorporating specialized fine-tunes for general capability, code, and reasoning.

Finetunes
3 models