Hunter 4B
A strong and smart model made to run on your normal laptop, built for agentic coding, real knowledge work, and cyber capabilities.
4B params / 262K–1M context
Full weightsOur latest agentic coding models, the strongest of their kinds.
Hunter 1.7B
Our smallest model, ready to be run on any device. Good for agentic coding, snappy reasoning, and smaller local cyber tasks.
1.7B params / 256K context
Full weightsCompare
How we eval every release| Hunter 4B | Hunter 1.7B |
|---|---|
| Parameters4B | Parameters1.7B |
| Context262K–1M | Context256K |
| Runs withllama.cpp, Ollama, LM Studio | Runs withHugging Face Transformers |
| LicenseApache-2.0 | LicenseApache-2.0 |
| Knowledge cutoffDec 2025 | Knowledge cutoffJun 2026 |
Run it locally
Release notesollama run hf.co/Lalakai/Hunter-1-4B:Q4_K_MWhat it needs
| Build | File | Memory |
|---|---|---|
| 4-bit | about 2.5 GB | about 3.5 GB at 4K context |
| FP16 | about 8 GB | about 10 GB RAM or VRAM |
Shipped as GGUF, so the same weights also load in llama.cpp and LM Studio. FP16 gives the best output quality if you have the memory.
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "Lalakai/Hunter-1-1.7B"
tokenizer = AutoTokenizer.from_pretrained(
model_id, trust_remote_code=True
)
model = AutoModelForCausalLM.from_pretrained(
model_id, trust_remote_code=True
)What it needs
- Python with Hugging Face Transformers
trust_remote_code=Truewhen loading- Setup steps and example code on the model card