OpenThinker

In the rapidly evolving landscape of artificial intelligence, OpenThinker has emerged as one of the most promising open-source models designed specifically for advanced reasoning. With models like OpenThinker-7B and OpenThinker-32B, this initiative is setting new standards in mathematical problem-solving, coding, and scientific analysis.

If you’re looking for an AI model that delivers exceptional performance in reasoning tasks while being fully open-source, you’ve come to the right place. In this post, we’ll break down everything you need to know about OpenThinker, from its unique features to its impressive benchmark performances.

---

Download and Install OpenThinker

While you can download AI models to run locally, doing so requires technical know-how and can be time-consuming. Here’s the easiest way to download almost any AI model in minutes.

1. Install OpenThinker on Windows

Ollama is a tool that lets you download almost any open-source AI model with just one simple command.

If you’re using macOS or Linux, you’ll find the respective options at the end of this guide.

• Windows Download Ollama for Windows
  1. Double-click the .exe file and follow the on-screen instructions.
  2. Confirm the installation by opening Command Prompt or PowerShell and typing ollama --version.

---

2. Pull an AI Model from Ollama’s Library

1. 1. Open your terminal or command prompt.
   2. Type one of the following commands:
      ollama run openthinker:7b
      ollama run openthinker:32b

3. Run a Model Locally

After downloading a model, starting it is as simple as executing the same command:

And just like that, you’ll have one of the top AI models running locally on your computer.

Other Options to Install OpenThinker

• macOS Download Ollama for macOS
  1. Unzip the file and run the installer in your Terminal (it usually includes a .sh script).
  2. That’s it! Open your Terminal and type ollama --version to verify the installation.
• Linux
  1. Open a terminal window.
  2. Run the following command:
     curl -fsSL https://ollama.com/install.sh | sh
  3. Verify the installation by typing:
     ollama --version

     You should see the current version of Ollama.

What is OpenThinker?

At its core, OpenThinker is a series of open-source AI models developed by the Open Thoughts team—a collaboration between Bespoke Labs and the DataComp community. These models are designed to handle complex reasoning tasks and are fine-tuned from Qwen2.5 models using the OpenThoughts-114k dataset.

Why is OpenThinker Important?

Unlike many proprietary models that require extensive resources, OpenThinker achieves state-of-the-art performance with significantly fewer training examples. For instance, OpenThinker-32B delivers superior results using only 114,000 training examples, compared to DeepSeek’s 800,000.

---

Key Features of OpenThinker AI Models

Feature	OpenThinker-7B	OpenThinker-32B
Base Model	Qwen2.5-7B-Instruct	Qwen2.5-32B-Instruct
Fine-Tuned On	OpenThoughts-114k dataset	OpenThoughts-114k dataset
Parameters	7.62 billion	32.8 billion
Training Hardware	4x 8xH100 nodes, 20 hours	8xH100 P5 nodes, 90 hours on 4 nodes
Context Length	16k tokens	16k tokens
Performance	Outperforms Bespoke-Stratos-7B	Best open-data reasoning model to date
License	Apache 2.0	Apache 2.0

---

Performance Benchmarks: How OpenThinker Stacks Up

The real measure of an AI model lies in its performance. Let’s examine how OpenThinker-7B and OpenThinker-32B fare against key benchmarks.

OpenThinker-7B Performance

Benchmark	Score
AIME24	31.3
MATH500	83.0
GPQA-Diamond	42.4
LCBv2 Easy	75.3
LCBv2 Medium	28.6
LCBv2 Hard	6.5
LCBv2 All	39.9

While it outperforms Bespoke-Stratos-7B, it still lags behind proprietary models like GPT-4 on some metrics. However, its open-source nature makes it a valuable alternative for researchers and developers who need a reliable reasoning model.

OpenThinker-32B Performance

Benchmark	Score
AIME24 I/II	66.0
AIME25 I	53.3
MATH500	90.6
GPQA Diamond	61.6
LCBv2	68.9

Notably, OpenThinker-32B’s 90.6 score on MATH500 and 61.6 on GPQA-Diamond make it the best open-data reasoning model currently available.

---

Open-Source Advantages: Why OpenThinker Stands Out

One of OpenThinker’s biggest strengths is its open-source nature, which provides several key advantages:

• Accessibility: Available to anyone without licensing fees or restrictions.
• Transparency: All datasets, training procedures, and model weights are publicly accessible.
• Community-Driven Innovation: Developers and researchers can contribute to and improve the model.

All resources, including model weights and code, can be found on platforms like Hugging Face and GitHub:

• Model Weights: Open Thoughts Hugging Face
• Datasets: OpenThoughts-114k
• Code: Open Thoughts GitHub
• Evaluation Tool: Evalchemy GitHub

Unlike proprietary models that remain closed-source, OpenThinker fosters an open and collaborative AI ecosystem.

---

Recent Developments: OpenThinker’s Growing Popularity

OpenThinker is gaining recognition in the AI community, with several notable updates:

• February 14, 2025: OpenThinker now features an online playground for interactive testing.
• February 13, 2025: OpenThinker is available on Ollama for easy local inference.
• February 12, 2025