vulture3/G4-WanaMeroQueen-v3-31B

VISIONPricing:Input $0.48 / Cached $0.1 / Output $1.44Concurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 4, 2026Architecture:Transformer Featherless Exclusive Cold

vulture3/G4-WanaMeroQueen-v3-31B is a 31 billion parameter merged language model, created by vulture3 using the DARE TIES method. It combines kawaimasa/Wanabi-Gemma4-31B, zerofata/G4-MeroMero-v2-31B, and aifeifei798/Gemma-4-Queen-31B-it. This model aims to improve Japanese language capabilities and logical coherence compared to previous merges, while maintaining a 32768 token context length.

Loading preview...

Overview

vulture3/G4-WanaMeroQueen-v3-31B is a 31 billion parameter language model resulting from a strategic merge of several pre-trained models. Developed by vulture3, this iteration specifically addresses observed degradation in Japanese language proficiency and logical flow in prior merges. The goal was to achieve a nuanced balance, preventing excessive deterioration of Japanese capabilities while enhancing the overall coherence of generated text.

Merge Details

This model was constructed using the DARE TIES merge method, with kawaimasa/Wanabi-Gemma4-31B serving as the base model. The merge incorporated the following models:

  • kawaimasa/Wanabi-Gemma4-31B
  • zerofata/G4-MeroMero-v2-31B
  • aifeifei798/Gemma-4-Queen-31B-it

Specific weighting and density parameters were applied to aifeifei798/Gemma-4-Queen-31B-it and zerofata/G4-MeroMero-v2-31B during the merge process to achieve the desired linguistic characteristics.

Key Capabilities

  • Improved Japanese Language Handling: Designed to mitigate the loss of Japanese language quality seen in previous merges.
  • Enhanced Coherence: Aims for more logical and consistent conversational flow.
  • 31 Billion Parameters: Offers substantial capacity for complex language tasks.
  • 32768 Token Context Length: Supports processing of extensive input sequences.

Should I use this for my use case?

This model is particularly suitable for applications requiring strong Japanese language generation and understanding, where maintaining logical consistency in dialogue or text is crucial. If your use case involves generating nuanced Japanese content or requires a model that balances multiple base model strengths without sacrificing specific language quality, G4-WanaMeroQueen-v3-31B is a strong candidate.

Finetunes
3 models