Project | SFT dataset | RL dataset | SFT model | RL model

OpenThinker-Agent-v1-SFT

OpenThoughts-Agent is an open-source effort to curate the best datasets for training agents. Our first release includes datasets, models and our research codebase.

OpenThinker-Agent-v1 is a model trained for agentic tasks such as Terminal-Bench 2.0 and SWE-Bench.

The OpenThinker-Agent-v1 model is post-trained from Qwen/Qwen3-8B. It is SFT-ed on the OpenThoughts-Agent-v1-SFT dataset, then RL-ed on the OpenThoughts-Agent-v1-RL dataset.

This OpenThinker-Agent-v1-SFT model is the model after the SFT stage. For the model after both SFT and RL stages, see OpenThinker-Agent-v1.

• Homepage: https://www.open-thoughts.ai/agent
• Repository: https://github.com/open-thoughts/OpenThoughts-Agent

OpenThinker-Agent-v1 Model Performance

Our OpenThinker-Agent-v1 model is the state-of-the-art model at its scale on agent benchmarks.

Data

We built OpenThinker-Agent-v1 in two stages: supervised fine-tuning, followed by reinforcement learning. Each stage required its own data pipeline – RL tasks (instructions, environments, and verifiers) and SFT traces from strong teacher agents completing tasks.

OpenThoughts-Agent-v1-SFT is an SFT trace dataset containing approximately 15,200 traces drawn from two different data sources we curate:

• nl2bash: Simple synthetically generated tasks where the agent has to format shell commands effectively
• InferredBugs: A set of bugs in C# and Java collected by Microsoft that we turned into tasks

We generate the traces using QuantTrio/GLM-4.6-AWQ and Terminus-2 agentic harness with a maximum of 32 turns. We use the default sampling parameters from vLLM and a maximum context length of 64K. Please see our example data generation script for more details.

OpenThoughts-Agent-v1-RL is an RL dataset containing ~720 tasks drawn from the nl2bash verified dataset.

To stabilize training, we built a three-stage filtration pipeline that prunes tasks before they ever hit the learner:

1. Bad verifiers filter: drop tasks with flaky or excessively slow verifiers.
2. Environment stability: remove tasks whose containers take too long to build or tear down. Optional difficulty filter: discard tasks that even a strong model (GPT-5 Codex) cannot solve in a single pass.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 4e-05
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 16
• total_train_batch_size: 16
• total_eval_batch_size: 128
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.98) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 7.0

Framework versions

• Transformers 4.56.0
• Pytorch 2.9.0+cu128
• Datasets 4.4.1
• Tokenizers 0.22.1

Links

• 🌐 OpenThoughts-Agent project page
• 💻 OpenThoughts-Agent GitHub repository
• 🧠 OpenThoughts-Agent-v1-SFT dataset
• 🧠 OpenThoughts-Agent-v1-RL dataset
• 🧠 OpenThoughts-TB-dev dataset
• 🤖 OpenThinker-Agent-v1 model
• 🤖 OpenThinker-Agent-v1-SFT model --> this model

Citation

@misc{openthoughts-agent,
  author = {Team, OpenThoughts-Agent},
  month = Dec,
  title = {{OpenThoughts-Agent}},
  howpublished = {https://www.open-thoughts.ai/blog/agent},
  year = {2025}
}