stgallenquants/OpenThinkerAgent-32B

TEXT GENERATIONPricing:Input $0.408 / Cached $0.0816 / Output $1.972Concurrent Unit Cost:2Model Size:32BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 28, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

OpenThinkerAgent-32B by stgallenquants is a 32 billion parameter language model, post-trained from Qwen3-32B with full-parameter SFT on a 100,000-example agentic dataset. This model is specifically optimized for agentic tasks, demonstrating strong performance across seven agentic benchmarks. With a context length of 32768 tokens, it is designed for complex, multi-turn agentic workflows.

Loading preview...

OpenThinkerAgent-32B: An Agentic Language Model

OpenThinkerAgent-32B is a 32 billion parameter model developed by stgallenquants, built upon the Qwen3-32B architecture. Its core differentiator lies in its specialized training for agentic tasks, making it highly effective in scenarios requiring autonomous decision-making and tool use.

Key Capabilities & Training

  • Agentic Optimization: Post-trained using full-parameter Supervised Fine-Tuning (SFT) on the extensive OpenThoughts-Agent-SFT-100K dataset, comprising 100,000 task-agent trajectory pairs.
  • High-Quality Data: The training data is sourced from top task sources like SWE-Smith and StackExchange, with trajectories generated by GLM-4.7-AWQ and filtered for traces with at least 5 model turns.
  • Performance Leader: Evaluated across a suite of seven agentic benchmarks, OpenThinkerAgent-32B consistently outperforms other open-data 32B models, achieving an average accuracy of 44.8%.
    • Notable scores include 55.7% on SWE-Bench-Verified-100 and 41.3% on OpenThoughts-TBLite in the Terminus-2 harness.
  • Robust Training: Utilizes a 32768 token context length and was trained with bf16 precision and DeepSpeed ZeRO-3 for 5 epochs.

Ideal Use Cases

This model is particularly well-suited for applications requiring advanced agentic capabilities, such as:

  • Automated Software Engineering: Tasks like code generation, debugging, and issue resolution (e.g., SWE-Bench).
  • Terminal-based Operations: Interacting with command-line interfaces and executing complex sequences of commands.
  • Knowledge-based Agents: Developing agents that can navigate and utilize information from various domains (e.g., MedAgentBench, FinanceAgent-Terminal).

For more details, refer to the OpenThoughts-Agent project page and the GitHub repository.

Finetunes
1 models