Instructions to use Lite-Coder/LiteCoder-Terminal-4b-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Lite-Coder/LiteCoder-Terminal-4b-sft with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Lite-Coder/LiteCoder-Terminal-4b-sft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Lite-Coder/LiteCoder-Terminal-4b-sft")
model = AutoModelForCausalLM.from_pretrained("Lite-Coder/LiteCoder-Terminal-4b-sft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use Lite-Coder/LiteCoder-Terminal-4b-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Lite-Coder/LiteCoder-Terminal-4b-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Lite-Coder/LiteCoder-Terminal-4b-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Lite-Coder/LiteCoder-Terminal-4b-sft

SGLang

How to use Lite-Coder/LiteCoder-Terminal-4b-sft with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Lite-Coder/LiteCoder-Terminal-4b-sft" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Lite-Coder/LiteCoder-Terminal-4b-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Lite-Coder/LiteCoder-Terminal-4b-sft" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Lite-Coder/LiteCoder-Terminal-4b-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Lite-Coder/LiteCoder-Terminal-4b-sft with Docker Model Runner:
```
docker model run hf.co/Lite-Coder/LiteCoder-Terminal-4b-sft
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

LiteCoder-Terminal-4b-sft

Paper | Code | Blog Post

LiteCoder-Terminal-4b-sft is part of our latest release on lightweight code agents. The model is fine-tuned from Qwen3-4B-Instruct-2507 on the LiteCoder-Terminal-SFT dataset.

Compared to our previous preview version, we scaled up the training data from under 1,000 samples to 11,255 trajectories, incorporating a broader task taxonomy and diverse agent scaffolds. With these updates, the model shows consistent improvements across Terminal Bench evaluations.

Released Artifacts

Date	Type	Link
2026/04/13	Model	LiteCoder-Terminal-30b-a3b-sft
2026/04/13	Model	LiteCoder-Terminal-4b-sft
2026/04/13	Dataset	LiteCoder-Terminal-SFT
2026/04/13	Dataset	LiteCoder-Terminal-World-Model-SFT
2026/04/13	Dataset	LiteCoder-Terminal-RL-preview
2026/04/13	Code	icip-cas/LiteCoder

Results

Terminal Bench 1.0 Performance

Model	Agent	pass@1	pass@4
LiteCoder-Terminal-30b-a3b-sft	Terminus 2	24.38%	40%
Qwen3-30B-A3B-Nex-N1	Openhands	18.44%	32.5%
LiteCoder-30a3b-Terminal-preview	Terminus 2	16.56%	27.5%
Qwen3-30B-A3B-Instruct	Terminus 2	16.56%	28.75%
LiteCoder-Terminal-4b-sft	Terminus 2	14.69%	28.75%
OpenThinker-Agent-v1	Terminus 2	11.25%	25%
LiteCoder-4b-Terminal-preview	Terminus 2	9.38%	20%
Qwen3-4B-Instruct	Terminus 2	6.25%	15%

Terminal Bench 2.0 Performance

Model	Agent	pass@1	pass@4
LiteCoder-Terminal-30b-a3b-sft	Terminus 2	12.36%	23.60%
Qwen3-30B-A3B-Nex-N1	Openhands	12.36%	23.60%
LiteCoder-30a3b-Terminal-preview	Terminus 2	6.18%	13.75%
Qwen3-30B-A3B-Instruct	Terminus 2	5.34%	11.24%
LiteCoder-Terminal-4b-sft	Terminus 2	4.78%	10.11%
OpenThinker-Agent-v1	Terminus 2	4.49%	10.1%
LiteCoder-4b-Terminal-preview	Terminus 2	4.78%	12.36%
Qwen3-4B-Instruct	Terminus 2	1.12%	3.37%

Terminal Bench Pro Performance

Model	Agent	pass@1
LiteCoder-Terminal-30b-a3b-sft	Terminus 2	31.5%
LiteCoder-30a3b-Terminal-preview	Terminus 2	22.0%
LiteCoder-Terminal-4b-sft	Terminus 2	21.5%
Qwen3-30B-A3B-Nex-N1	Openhands	21.0%
Qwen3-30B-A3B-Instruct	Terminus 2	20.5%
OpenThinker-Agent-v1	Terminus 2	19.5%
LiteCoder-4b-Terminal-preview	Terminus 2	15.0%
Qwen3-4B-Instruct	Terminus 2	3.5%

Citation

@article{peng2026litecoderterminal,
  title={LiteCoder-Terminal: Scaling Long-Horizon Terminal Environments for Learning Language Agents},
  author={Peng, Xiaoxuan and Zhang, Kaiqi and Lu, Xinyu and Cao, Boxi and Lu, Yaojie and Lin, Hongyu and Han, Xianpei and Sun, Le},
  journal={arXiv preprint arXiv:2605.29559},
  year={2026}
}