Mert Yuksekgonul*, Daniel Koceja*, Xinhao Li*, Federico Bianchi*
Jed McCaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou†, Carlos Guestrin†, Yu Sun*

Stanford · NVIDIA · Astera Institute · UC San Diego · Together AI

TTT-Discover performs reinforcement learning at test time, allowing the LLM to continue training with experience specific to the problem at hand. We achieve new state-of-the-art across mathematics, GPU kernels, algorithms, and biology.

pip install -r requirements/requirements-math.txt

Set environment variables:

export TINKER_API_KEY="..."      
export WANDB_API_KEY="..."       
export WANDB_ENTITY="..."        

Task-specific requirements:

• GPU kernels: requirements/requirements-gpumode.txt
• AtCoder: requirements/requirements-ale.txt
• Denoising: requirements/denoising/requirements-denoising.txt (see README)

Requires SLURM. Launch AC1 (autocorrelation inequality) on 4 nodes:

python main_tinker_submitit.py \
    --nodes 4 \
    --partition default \
    --cpus-per-task 100 \
    env=ac1 \
    model_name="openai/gpt-oss-120b" \
    sampler_type=puct_backprop \
    initial_exp_type=random \
    num_epochs=50 \
    wandb_project="my-project" \
    wandb_name="ac1-run-1"

Or use a preconfigured script:

bash scripts/tinker/ac1.sh

See docs/launching.md for all parameters and docs/intro.md for adding new tasks.

This work builds on several outstanding projects and communities:

• GPU Mode — Community for GPU kernel optimization and the TriMul competition
• ALE-Bench — AtCoder-based benchmark for LLM evaluation
• AlphaEvolve — DeepMind's evolutionary coding agent
• OpenEvolve — Open-source implementation of AlphaEvolve
• Tinker — LLM training recipes and RL framework

@article{ttt-discover2026,
  title   = {Learning to Discover at Test Time},
  author  = {Yuksekgonul, Mert and Koceja, Daniel and Li, Xinhao 
             and Bianchi, Federico and McCaleb, Jed and Wang, Xiaolong 
             and Kautz, Jan and Choi, Yejin and Zou, James 
             and Guestrin, Carlos and Sun, Yu},
  journal = {arXiv preprint arXiv:2601.16175},
  year    = {2026}
}

This project is licensed under the MIT License - see the LICENSE file for details.