BlueNipples/Anansi-35B-A3B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 19, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

BlueNipples/Anansi-35B-A3B is a 35.1 billion parameter language model developed by BlueNipples, designed to balance instruct following, narrative reasoning, and prose quality. It was created through a DARE-TIES merge of two base models, followed by a targeted LM head tensor interpolation. This model excels at complex instruction following while maintaining concise, character-focused narrative reasoning and enhanced prose, making it suitable for creative writing and revision tasks.

Loading preview...

Anansi-35B-A3B: A Merged Model for Narrative and Instruction

Anansi-35B-A3B is a 35.1 billion parameter model developed by BlueNipples, specifically engineered to provide a strong balance of instruct following, narrative reasoning, and high-quality prose. It is designed to be efficient enough for use on systems with more limited resources.

Key Capabilities & Features

  • Balanced Performance: Achieves a good equilibrium between following complex instructions, generating narrative-based reasoning, and producing refined prose.
  • Merge Architecture: Created using a 35/65 DARE-TIES merge of huihui-ai/Huihui-Agents-A1-abliterated (for instruction following) and Gryphe/WorldSim-Opus-3.6-35B-A3B (for concise narrative reasoning and prose).
  • Prose Enhancement: Features a 30% LM head tensor interpolation with ReadyArt/Melody1437-35B-A3B to graft on superior prose quality, while preserving underlying logic and instruct capabilities.
  • Efficient Revision: Capable of reliably performing tasks like prose refinement and revision in a single pass, making it suitable for creative writing workflows.

Ideal Use Cases

  • Creative Writing: Generating narrative content, character-focused reasoning, and improving prose style.
  • Instruction Following: Handling complex and multi-step instructions effectively.
  • Content Revision: Assisting with refining and revising written text, particularly for stylistic improvements.
  • Resource-Constrained Environments: Optimized for users with 8GB VRAM or similar 'potato' setups, offering strong performance without requiring high-end hardware.

Finetunes
3 models