RicardoEstep/RPBizkit-v11-12B-Thinking

TEXT GENERATIONPricing:Input $0.87 / Cached $0.2 / Output $0.99Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 6, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

RicardoEstep/RPBizkit-v11-12B-Thinking is a 12-billion parameter experimental language model based on a "Mistral Nemo" mix, created by RicardoEstep using a DELLA merge with Mergekit. This model is designed for general text generation and processing, leveraging its mixed architecture for potentially enhanced reasoning capabilities. It features a 32768-token context window, with reported support for up to 1M context, and is noted for its "uncensored" nature, making it suitable for diverse and unrestricted content generation tasks.

Loading preview...

RicardoEstep/RPBizkit-v11-12B-Thinking Overview

This model is an experimental 12-billion parameter language model developed by RicardoEstep, utilizing a "Mistral Nemo" mix. It was created using a DELLA merge with Mergekit, combining RicardoEstep/RPBizkit-v11-12B-Base with DavidAU/Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC. The merge configuration specifically used a della method with a density of 0.4 and a weight of 0.8 for the secondary model, aiming for "thinking" deltas.

Key Capabilities

  • Experimental Architecture: Leverages a unique DELLA merge of Mistral-Nemo based models for potentially distinct performance characteristics.
  • Extended Context Window: Features a clean tokenizer and embedding sizes (131072), supporting a 32768-token context window, with reported support for up to 1M context.
  • "Uncensored" Nature: Designed with "uncensored" base models, allowing for broader and less restricted content generation.
  • Multilingual Adaptability: While primarily English, it can be prompted to "think" in other languages by forcing a system prompt to analyze context first.

Good For

  • General Text Generation: Suitable for a wide range of text generation tasks where an experimental and less constrained model is desired.
  • Exploratory AI Development: Ideal for researchers and developers interested in testing models with unique merging techniques and "uncensored" characteristics.
  • Content Creation: Can be used for creative writing, roleplay, and other applications requiring flexible and unrestricted output, provided appropriate system prompts are used for non-English content.

Rapid benchmarking on a quantized version (Q8_0) with an 8K context window showed scores of 56.25% acc and 76.56% acc_norm on Hellaswag, and varying scores on ifeval v4.

Finetunes
2 models