unsloth/Phi-4-reasoning-plus

TEXT GENERATIONConcurrency Cost:1Model Size:14.7BQuant:FP8Ctx Length:32kPublished:May 1, 2025License:mitArchitecture:Transformer0.0K Open Weights Cold

Phi-4-reasoning-plus is a 14.7 billion parameter decoder-only Transformer model developed by Microsoft Research, fine-tuned from Phi-4. It is optimized for advanced reasoning tasks in math, science, and coding, utilizing supervised fine-tuning on chain-of-thought traces and reinforcement learning. With a 32k token context length, this model excels at generating detailed reasoning chains followed by summarized solutions. Its primary use cases include accelerating research in language models and serving as a building block for generative AI applications requiring strong reasoning capabilities in memory/compute-constrained and latency-bound environments.

Loading preview...

What is unsloth/Phi-4-reasoning-plus?

unsloth/Phi-4-reasoning-plus is a 14.7 billion parameter language model from Microsoft Research, built upon the Phi-4 architecture. It is specifically fine-tuned for advanced reasoning tasks in mathematics, science, and coding. The model leverages supervised fine-tuning on chain-of-thought (CoT) traces and reinforcement learning to enhance its problem-solving abilities.

Key Capabilities

  • Enhanced Reasoning: Excels in complex reasoning tasks, generating detailed thought processes before providing solutions.
  • Specialized Training: Fine-tuned on high-quality datasets focusing on math, science, and coding skills.
  • Context Length: Supports a substantial 32k token context window, with experimental support up to 64k tokens for longer reasoning sequences.
  • Performance: Demonstrates strong performance on reasoning benchmarks like AIME, OmniMath, and GPQA-Diamond, often outperforming larger open-weight models.
  • Structured Output: Designed to produce responses with distinct 'Thought' and 'Solution' sections, aiding clarity and analysis.

Good for

  • Research & Development: Ideal for accelerating research in language models and as a foundation for generative AI features.
  • Reasoning-Intensive Applications: Suited for applications requiring strong logical deduction, problem-solving, and multi-step reasoning.
  • Resource-Constrained Environments: Optimized for use in memory/compute-constrained and latency-bound scenarios due to its efficient architecture.
  • Educational Tools: Can be used to develop tools that explain complex concepts through detailed reasoning steps.