biennequants/OpenThinker-7B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 7, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

OpenThinker-7B is a 7.6 billion parameter instruction-tuned language model developed by biennequants, fine-tuned from Qwen/Qwen2.5-7B-Instruct. It is specifically optimized for reasoning tasks, leveraging the OpenThoughts-114k dataset which distills DeepSeek-R1. This model demonstrates improved performance on benchmarks like AIME24, MATH500, and GPQA-Diamond compared to its predecessor, making it suitable for complex analytical and problem-solving applications.

Loading preview...

OpenThinker-7B: A Reasoning-Optimized Language Model

OpenThinker-7B is a 7.6 billion parameter model fine-tuned from Qwen/Qwen2.5-7B-Instruct, developed by biennequants. Its core differentiator is its optimization for complex reasoning tasks, achieved through fine-tuning on the OpenThoughts-114k dataset. This dataset is derived by distilling DeepSeek-R1, focusing on high-quality reasoning examples.

Key Capabilities & Performance

OpenThinker-7B shows notable improvements in reasoning benchmarks compared to previous models like Bespoke-Stratos-7B. Evaluated using the open-source tool Evalchemy, it achieves:

  • AIME24: 31.3 (vs. 22.7 for Bespoke-Stratos-7B)
  • MATH500: 83.0 (vs. 79.6 for Bespoke-Stratos-7B)
  • GPQA-Diamond: 42.4 (vs. 38.9 for Bespoke-Stratos-7B)

This model is part of a fully open-source initiative, with its model weights, datasets, data generation code, and training code all publicly available. A detailed paper on OpenThoughts is also available here.

When to Use This Model

OpenThinker-7B is ideal for applications requiring strong analytical and problem-solving capabilities, particularly in domains tested by benchmarks like AIME24 and MATH500. Its open-source nature also makes it suitable for researchers and developers looking to build upon or further fine-tune reasoning-focused models.