GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
TobiasLogic
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
TobiasLogic
Model Tree
Input Modalities
Output Modalities
Supported Functionality
Trained on 1,000,000 positions sampled from random play and labeled with
Stockfish's best move (single-move samples of the form 1.<san>). The arena
queries a model with a constant "1." prompt and constrains the output to the
legal moves of the current position (via outlines), so this model learns a
distribution over strong move shapes rather than board-specific tactics.
transformers causal LM; drop-in for the arena."1."), so no model in that arena — including
this one — can calculate tactics. Expect casual, not strong, play.python
from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("TobiasLogic/gambit-gpt")model = AutoModelForCausalLM.from_pretrained("TobiasLogic/gambit-gpt")
Or enter TobiasLogic/gambit-gpt directly in the arena Space.
v2 will train on master-game openings for a stronger move prior. A board-aware variant (that actually sees the position) is planned for real UCI play.