MagenticBrain
MagenticBrain is a 14B-parameter text-generation orchestration model from Microsoft Research AI Frontiers, built on Qwen/Qwen3-14B. It is supervised fine-tuned on agentic data and further trained with reinforcement learning on in-house terminal tasks. It plans multi-step tasks, selects declared tools, chains tool calls, and delegates to sub-agents, while not executing actions itself; real-world side effects occur in a host harness. It uses structured JSON tool calls, a submit-to-terminate protocol, and a 32K context. Weights are released under MIT; it is co-designed with and most thoroughly evaluated inside MagenticLite.
At a glance
- Parameters
- 14.77B
- Architecture
- qwen3
- Context length
- 40,960
- License
- mit
- Disk space
- 55 GB
- Software
- vLLM, Transformers
Will it run on your machine?
Save your machine to see a personalized rating and its reasoning.
Add my hardwareBest for
- Planning multi-step tasks
- Structured JSON tool selection and multi-turn tool chaining
- Sub-agent delegation to Fara1.5-9B for browser and desktop control
- Orchestration within the MagenticLite harness
- Agentic workflow coordination using declared tools
Tradeoffs
- Does not execute actions itself; side effects require a host harness
- Not a general-purpose chat model
- Selects only from tools declared at inference time and does not invent new tools
- Inference code, training recipes, and execution harness are not included in the weights release
- Most thoroughly evaluated in the MagenticLite configuration
Ways to run it
vLLM · Linux
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
vllm serve microsoft/MagenticBrain --enable-auto-tool-choice --tool-call-parser hermes --max-model-len 32768Transformers · See official guide
Repository-specific command in publisher documentation; confirm dependencies and hardware in the source.
import json
from transformers import AutoTokenizer, AutoModelForCausalLM
model_name = "microsoft/MagenticBrain"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto",
)
tools = [
{
"type": "function",
"function": {
"name": "read_file",
"description": "Read the contents of a file at the given path.",
"parameters": {
"type": "object",
"properties": {"path": {"type": "string"}},
"required": ["path"],
},
},
},
{
"type": "function",
"function": {
"name": "submit",
"description": "Signal that the task is complete.",
"parameters": {"type": "object", "properties": {}},
},
},
]
messages = [
{"role": "system", "content": "You are an orchestration agent. Plan steps and call only the tools declared below."},
{"role": "user", "content": "Summarize the contents of /tmp/report.md."},
]
text = tokenizer.apply_chat_template(
messages,
tools=tools,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False, # thinking is disabled by default for MagenticBrain
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
output_ids = model.generate(**inputs, max_new_tokens=1024)[0][inputs.input_ids.shape[1]:]
print(tokenizer.decode(output_ids, skip_special_tokens=True))Explore its uses
Keep exploring
Bonsai 2 · 27B
A compressed 27B-class reasoning model with publisher-provided GGUF packs and a dedicated llama.cpp fork for CUDA, Metal, and CPU.
DeepSeek V4.1 Flash
An open-weight model for reasoning, coding, agents, and image understanding. DeepSeek V4.1 Flash combines a 552B backbone with sparse activation and up to one million tokens of context. Choose the original checkpoint or a community uncensored variant below.
Qwen3.8
Explore 4 versions of Qwen3.8. Compare model sizes, packages, hardware requirements, and available setups.