microsoft/X-Reasoner-7B

VISIONPricing:Input $0.4 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:32kPublished:Feb 3, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

microsoft/X-Reasoner-7B is a 7 billion parameter vision-language model developed by Microsoft, post-trained on general-domain text for enhanced reasoning capabilities. It utilizes a two-stage approach involving supervised fine-tuning with distilled long chain-of-thoughts and reinforcement learning with verifiable rewards. This model excels at transferring reasoning to both multimodal and out-of-domain settings, outperforming existing state-of-the-art models on various general and medical benchmarks. With a context length of 32768 tokens, it is designed for generalizable reasoning across modalities and domains.

Loading preview...

Overview

X-Reasoner-7B is a 7 billion parameter vision-language model developed by Microsoft, specifically engineered for generalizable reasoning across different modalities and domains. Unlike many models, it is post-trained exclusively on general-domain text, focusing on robust reasoning transfer.

Key Capabilities

  • Advanced Reasoning: Achieves strong reasoning performance through a novel two-stage training process.
  • Multimodal Understanding: Successfully transfers reasoning capabilities to multimodal inputs, integrating vision and language.
  • Domain Generalization: Demonstrates superior performance in out-of-domain scenarios, including general and medical benchmarks, compared to models trained with in-domain data.
  • Chain-of-Thought Integration: Leverages distilled long chain-of-thoughts during supervised fine-tuning to enhance reasoning depth.
  • Reinforcement Learning: Incorporates reinforcement learning with verifiable rewards for further performance optimization.

When to Use This Model

This model is particularly well-suited for applications requiring:

  • Complex Reasoning Tasks: Where the ability to process and reason over diverse information is critical.
  • Multimodal AI Systems: For tasks that involve understanding and reasoning from both visual and textual data.
  • Domain-Agnostic Applications: When a model needs to perform reliably across various fields without extensive in-domain fine-tuning.
  • Medical AI: Shows strong performance on medical benchmarks, indicating potential for healthcare applications requiring reasoning.

Finetunes
1 models