tmax-9b

Image: image

💻 Code · 🤗 Models & Data · 📜 Paper · 📓 Blog

Note. For full information, go check out the Tmax paper here.

TMax 9B#

TMax 9B is a model trained using DPPO on top of Qwen 3.5 9B for use as a terminal-agent. It achieves roughly 27% on Terminal Bench 2.0 after 200 steps of RL training, improving by ~6 points over Qwen 3.5 9B with our harness.

Image: image

This model is part of a collection of terminal agents in various sizes.

Additionally, we provide model checkpoints as branches of the repository. The main model checkpoint is step 200 as this performed best on TBLite.

We also provide all model rollouts and logprobs over the course of training here! This allows you to inspect what went on during our runs. See the folder for more details (I recommend pointing an agent at it to make plots, etc).

Evaluation Results#

For details on evaluation methodology please check our paper. In general, we used a podman (docker) backend with default timeouts and custom harness similar to mini-swe-agent. For the 'daytona' runs, we used the daytona backend. For Lite/2.1, we show mean and standard error over 3 runs. For daytona, we show it over 5 runs.

Model Details#

Model Description#

• Developed by: Ai2
• Language(s) (NLP): English
• License: Apache 2.0
• Finetuned from model [optional]: Qwen 3.5 9B
• Dataset: TMax-15k

Use#

To use this model, we recommend serving with vllm (or your inference framework of choice) with:

uvx vllm==0.19.1 serve allenai/tmax-9b \
  --served-model-name tmax-9b \
  --enable-auto-tool-choice \
  --tool-call-parser qwen3_xml \
  --port 8008 \
  --max-model-len 65536 \
  --tensor-parallel-size 8 \
  --language_model_only

Make sure to set language_model_only as we removed the vision head during training.

For more details on evaluation, please see our codebase.

Hyperparameters#

This model was trained using DPPO with the following hyperparameters:

• base model: hamishivi/Qwen3.5-9B
• Dataset: tmax 15K
• Max prompt tokens: 2048
• Max per-turn tokens: 16384
• Max overall tokens: 65536
• Pack length: 67584
• Per-device train batch size: 1
• Unique prompts per rollout: 8
• Samples per prompt rollout: 32
• Async steps: 4
• Max steps: 64
• Learning rate: 1e-6
• LR scheduler: constant
• Total training steps: 500 steps (this checkpoint is from 200 steps of training, which performed best on TBLite)
• Sampling Temperature: 1.0
• KL Beta: 0.0
• Loss fn: DPPO
• Divergence: binary TV
• TV threshold: 0.1
• Advantage normalization: centered (no division by stdev)
• FP32 LM head: true

For more details on training, please see our codebase.

License#

This model is licensed under Apache 2.0. It is intended for research and educational use in accordance with Ai2's Responsible Use Guidelines.

Citation#

If you use our model or data, please cite our paper:

@misc{ivison2026tmaxsimplerecipeterminal,
      title={Tmax: A simple recipe for terminal agents}, 
      author={Hamish Ivison and Junjie Oscar Yin and Rulin Shao and Teng Xiao and Nathan Lambert and Hannaneh Hajishirzi},
      year={2026},
      eprint={2606.23321},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2606.23321}, 
}