Falcon-H1R-7B

This repository presents Falcon-H1R-7B, a reasoning-specialized model introduced in the paper Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling.

Built on top of Falcon-H1-7B-Base, it was trained via cold-start supervised fine-tuning with long reasoning traces and further enhanced by scaling RL with GRPO. The model demonstrates outstanding performance across various benchmark evaluations, including mathematics, programming, instruction following, and general logic.

Model Description

• Developed by: Technology Innovation Institute
• Model type: Causal decoder-only
• Architecture: Hybrid (Transformers + Mamba2) architecture
• Language(s): English, Multilingual
• License: Falcon-LLM License

Training details

For more details about the training protocol of this model, please refer to the Falcon-H1R technical blogpost and Technical Report.

Usage

Currently to use this model, you can either rely on Hugging Face transformers, vLLM or SGLang library.

Inference

Make sure to install the latest version of transformers or vLLM or SGLang.

pip install transformers
pip install mamba-ssm[causal-conv1d]

For vLLM, make sure to install vllm=0.11.0:

pip install "vllm>=0.11.0"

Sampling Parameters

We recommend using a temperature of 0.6 and top-p as 0.95 with max new tokens up to 65536. For supported frameworks, you can adjust the repetition_penalty and presence_penalty parameters to reduce endless repetitions. For reasoning tasks with continuous batching and requiring higher max new tokens, we recommend to use TP=2.

🤗 Transformers

Refer to the snippet below to run H1R models using 🤗 transformers. Model will generate think content wrapped in a <think>...</think> block, followed by the final response.

from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = "tiiuae/Falcon-H1R-7B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", dtype="auto")

messages = [
    {"role": "user", "content": "What is the derivative of x^2?"},
]
inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt")

outputs = model.generate(
    inputs.to(model.device),
    max_new_tokens=65536,
    temperature=0.6,
    top_p=0.95,
    do_sample=True,
)
print(tokenizer.decode(outputs[0]))

vLLM

For vLLM, simply start a server by executing the command below:

vllm serve tiiuae/Falcon-H1R-7B \
  --tensor-parallel-size 1 \
  --data-parallel-size 1 \
  --reasoning-parser deepseek_r1

• You can reduce --max-model-len to preserve memory. Default value is 262144 which is quite large but not necessary for most scenarios.
• For function calling, append --enable-auto-tool-choice and --tool-call-parser hermes to the vllm serve command.

vLLM client execution code:

from openai import OpenAI
import json

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="EMPTY",
)

completion = client.chat.completions.create(
    model="tiiuae/Falcon-H1R-7B",
    messages=[
        {"role": "user", "content": "If the product of two numbers is 360 and their GCD is 6, what is their LCM?"},
    ],
    temperature=0.6,
    top_p=0.95,
    max_tokens=65536
)

msg = completion.choices[0].message

print(json.dumps({
    "reasoning": msg.reasoning_content,
    "answer": msg.content
}, indent=2))

SGLang

For SGLang, simply start a server by executing the command below:

python -m sglang.launch_server \
  --model tiiuae/Falcon-H1R-7B \
  --tensor-parallel-size 1 \
  --reasoning-parser deepseek-r1

SGLang client execution code:

from openai import OpenAI
import json

client = OpenAI(
    base_url="http://localhost:30000/v1",
    api_key="EMPTY",
)

completion = client.chat.completions.create(
    model="tiiuae/Falcon-H1R-7B",
    messages=[
        {"role": "user", "content": "How many solutions in integers satisfy ∣x∣+∣y∣=20?"},
    ],
    max_tokens=65536,
    temperature=0.6,
    top_p=0.95,
)

msg = completion.choices[0].message

print(json.dumps({
    "reasoning": msg.reasoning_content,
    "answer": msg.content
}, indent=2))

Evaluation

Falcon-H1R achieves state of art results in reasoning benchmarks.

TTS represents test time scaling results on few of the benchmarks that we evaluated via DeepConf. Note that AMO-Bench* is limited to the parser-verifiable subset which comprises 39 problems.

Useful links

• View our release blogpost.
• View our technical report.
• Feel free to join our discord server if you have any questions or to interact with our researchers and developers.

Citation

If the Falcon-H1R family of reasoning models is helpful to your work, feel free to give us a cite.

@misc{falcon-h1r,
      title={Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling}, 
      author={Falcon LLM Team and Iheb Chaabane and Puneesh Khanna and Suhail Mohmad and Slim Frikha and Shi Hu and Abdalgader Abubaker and Reda Alami and Mikhail Lubinets and Mohamed El Amine Seddik and Hakim Hacid},
      year={2026},
      eprint={2601.02346},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2601.02346}, 
}