PocketPlaning-1 is a specialized AI agent planning model built on Qwen3.5-27B. It is fine-tuned for intelligent task routing, complexity estimation, and tool-calling orchestration in agentic workflows.

Key Capabilities

Task Planning: Decomposes complex tasks into ordered subtasks with dependency tracking
Complexity Routing: Classifies subtasks as EASY (handle locally) or HARD (escalate to frontier model)
Tool Calling: Native function-calling support for agentic tool orchestration
Bilingual: Full support for English and Chinese

Model Details

Attribute	Value
Base Model	Qwen3.5-27B
Parameters	27B
Context Length	262,144 tokens
Architecture	Qwen3.5ForConditionalGeneration
Precision	bfloat16
License	Qwen License

Usage

With Transformers

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "PocketBrains/PocketPlaning-1"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="bfloat16",
    device_map="auto",
    trust_remote_code=True,
)

messages = [{"role": "user", "content": "Help me write a Python script to parse CSV files"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=2048)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

With vLLM

bash

vllm serve PocketBrains/PocketPlaning-1 \
    --tensor-parallel-size 2 \
    --max-model-len 131072 \
    --dtype bfloat16 \
    --trust-remote-code \
    --enable-auto-tool-choice \
    --tool-call-parser qwen3_coder

Intended Use

PocketPlaning-1 is designed to serve as the planning and routing layer inside AI agent systems. It determines what to do next and how much compute to spend doing it, enabling cost-efficient agent execution by routing simple tasks locally and escalating complex tasks to frontier models.

Limitations

This model is optimized for agent planning and tool-calling scenarios; it is not a general-purpose chatbot.
Performance is best when used within an agentic framework with tool definitions provided.
The model inherits limitations from the base Qwen3.5-27B model.

Citation

bibtex

@misc{pocketplan2026,
  title={PocketPlaning-1: Efficient Planning Models for AI Agents},
  author={PocketBrains Inc.},
  year={2026},
  url={https://huggingface.co/PocketBrains/PocketPlaning-1}
}

About PocketBrains

PocketBrains Inc. builds open-source efficient models for AI agents. We train specialized small models that make frontier model usage economically viable by intelligently routing tasks based on complexity.

Website: pocketbrainsai.com
Contact: founder@pocketbrainsai.com

PocketBrains233/PocketPlanning-1

PocketPlaning-1