Cannae-AI/Gemini-3.1-Pro-Qwen3-14B

TEXT GENERATIONConcurrency Cost:1Model Size:14BQuant:FP8Ctx Length:32kPublished:Mar 22, 2026License:mitArchitecture:Transformer0.0K Open Weights Cold

Gemini-3.1-Pro-Qwen3-14B by Cannae-AI is a 14 billion parameter fine-tuned model based on the Qwen3 architecture, specifically designed for expert-level reasoning tasks. It leverages Gemini 3.1 Pro's advanced capabilities to enhance analytical depth, logical coherence, and the ability to synthesize conflicting information. This model excels in multi-step reasoning, derivation, and complex problem synthesis, making it ideal for research and applications requiring advanced analytical capabilities. It has a context length of 32768 tokens.

Loading preview...

Overview

Cannae-AI's Gemini-3.1-Pro-Qwen3-14B is a 14 billion parameter model fine-tuned for expert-level reasoning tasks. This model significantly enhances analytical depth, logical coherence, and the ability to synthesize conflicting information by leveraging the advanced capabilities of Gemini 3.1 Pro.

Key Capabilities

  • Advanced Reasoning: Excels in multi-step reasoning, derivation, and complex problem synthesis.
  • High-Complexity Problem Solving: Designed to tackle expert-level problems requiring multi-step reasoning across logic, mathematics, and domain-specific areas.
  • Synthetic Data Training: Fine-tuned using a synthetic, high-complexity reasoning corpus, with Gemini 3.1 Flash as the prompting agent and Gemini 3.1 Pro as the solving agent.

Intended Use Cases

  • Advanced Reasoning Applications: Ideal for applications demanding sophisticated problem-solving.
  • AI Cognition Research: Suitable for research into multi-step logical reasoning.

Limitations

  • Performance in real-world scenarios may vary as it's trained on synthetic data.
  • Specialized focus on reasoning may lead to reduced performance in general NLP or casual conversation tasks.