---
library_name: transformers
license: other
license_name: nvidia-open-model-license
license_link: >-
https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/
pipeline_tag: text-generation
language:
- en
tags:
- nvidia
- nemotron-cascade-2
- reasoning
- general-purpose
- SFT
- RL
---
# Nemotron-Cascade-2-30B-A3B
[](https://arxiv.org/abs/2603.19220)
[](https://huggingface.co/datasets/nvidia/Nemotron-Cascade-2-SFT-Data)
[](https://huggingface.co/datasets/nvidia/Nemotron-Cascade-2-RL-data)
[](https://huggingface.co/collections/nvidia/nemotron-cascade-2)
## Introduction
We're excited to introduce [Nemotron-Cascade-2-30B-A3B](https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B), an open 30B MoE model with 3B activated parameters that delivers strong reasoning and agentic capabilities. It is post-trained from the [Nemotron-3-Nano-30B-A3B-Base](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16). Nemotron-Cascade-2-30B-A3B achieves ***gold medal*** performance in both the 2025 International Mathematical Olympiad (IMO) and the International Olympiad in Informatics (IOI). It operates in both **thinking** and **instruct** (non-thinking) modes.
## Benchmark Results
| Benchmark |
Nemotron-3-Nano-30B-A3B |
Nemotron-3-Super-120B-A12B |
Qwen3.5-35B-A3B |
Nemotron-Cascade-2-30B-A3B |
| Math |
| IMO 2025 |
- | - | - | 🏅 35 pts |
| IMO AnswerBench |
70.4‡ | 77.2‡ | 74.8‡ | 79.3 |
| IMO ProofBench |
- | - | - | 72.9 |
| AIME 2025 |
89.1 | 90.2 | 91.9‡ | 92.4 (98.6)† |
| AIME 2026 |
89.9‡ | 89.8‡ | 91.1‡ | 90.9 (95.0)† |
| HMMT Feb25 |
84.6‡ | 93.7 | 89.0 | 94.6 |
| Code Reasoning |
| IOI 2025 |
- | - | 348.6‡ | 🏅 439.3 |
| ICPC World Finals 2025 |
- | - | - | 🏅 10/12 |
| LiveCodeBench v6 (2408-2505) |
68.3 | 78.7 | 74.6 | 87.2 (88.4)† |
| LiveCodeBenchPro 25Q2 (Easy) |
54.5‡ | 81.7‡ | 81.1‡ | 87.0 (89.3)† |
| LiveCodeBenchPro 25Q2 (Med) |
3.50‡ | 23.2‡ | 17.8‡ | 27.6 (36.8)† |
| SciCode |
33.3 | 42.1 | 38.0 | 36.4 |
| Knowledge & STEM |
| MMLU-Redux |
- | - | 93.3 | 86.3 |
| MMLU-Pro |
78.3 | 83.7 | 85.3 | 79.8 |
| GPQA-Diamond |
73.0 | 79.2 | 84.2 | 76.1 |
| HLE (no tool) |
10.6 | 18.3 | 22.4 | 17.7 |
| Alignment & Instruction Following |
| ArenaHard v2 (Avg.) |
67.7 | - | 65.4‡ | 83.5 |
| – Hard Prompt |
72.1 | 73.9 | 64.5‡ | 88.2 |
| – Creative Writing |
63.2 | - | 66.3‡ | 78.7 |
| IFBench (prompt) |
71.5 | 72.6 | 70.2 | 82.9 |
| Scale AI Multi-Challenge |
38.5 | 55.2 | 60.0 | 45.3 |
| Long Context & Context Learning |
| AA-LCR |
35.9 | 58.3 | 58.5 | 39.1 |
| LongBench v2 |
39.6 | - | 59.0 | 40.3 |
| NIAH@1M (RULER Subset) |
94.8 | 98.3 | 94.3‡ | 99.0 |
| CL-Bench |
12.0‡ | - | 15.5‡ | 12.2 |
| Agentic |
| BFCL v4 |
53.8 | - | 67.3 | 52.9 |
| 𝜏²-Bench |
49.0 | 61.2 | 81.2 | 58.9 |
| Terminal Bench 2.0 |
8.5 | 31.0 | 40.5 | 21.1 |
| SWE Verified (OpenHands) |
38.8 | 60.5 | 69.2 | 50.2 |
| Multilingual |
| MMLU-ProX |
59.5 | 79.4 | 81.0 | 72.5 |
| WMT24++ (en -> xx) |
86.2 | 86.7 | 87.6‡ | 84.1 |
* † Numbers in brackets refers to Tool-Integrated Reasoning (TIR) results.
* ‡ For the baseline models, we evaluate them using the recommended settings when official numbers are unavailable; otherwise, we report the official numbers directly.
## Quick Start
- Nemotron-Cascade-2-30B-A3B follows the ChatML template and supports both thinking and instruct (non-thinking) modes. Reasoning content is enclosed within `` and `` tags. To activate the instruct (non-thinking) mode, we prepend `` to the beginning of the assistant’s response.
- Nemotron-Cascade-2-30B-A3B supports up to a 1M-token context length.
- Nemotron-Cascade-2-30B-A3B does not currently support OpenCode; it primarily supports OpenHands for agentic coding and SWE tasks.
- To reduce the context length in a multi-turn conversation, when the previous user turn involves thinking mode, only the final summary of the model's output will be added to the conversation history.
- Note that we do not define a separate `tool` role for tool responses; instead, we place them under the `user` role and warp them with `` and ``.
- We recommend setting the sampling parameters to temperature = 1.0 and top_p = 0.95.
### vLLM setup
Requires vLLM version >= 0.17.1. The following will create API endpoints at `http://localhost:8000/v1`:
- **Standard version**: Use the following command to create an API endpoint with a maximum context length of 1M tokens.
```shell
vllm serve nvidia/Nemotron-Cascade-2-30B-A3B --port 8000 --tensor-parallel-size 1 --gpu-memory-utilization 0.9 --max-model-len 262144 --reasoning-parser nemotron_v3 --mamba-ssm-cache-dtype float32 --port 8000 --trust_remote_code
```
- **Tool Call**: Use the following command to enable tool support.
```shell
vllm serve nvidia/Nemotron-Cascade-2-30B-A3B --port 8000 --tensor-parallel-size 1 --gpu-memory-utilization 0.9 --max-model-len 262144 --reasoning-parser nemotron_v3 --mamba-ssm-cache-dtype float32 --port 8000 --trust_remote_code --enable-auto-tool-choice --tool-call-parser qwen3_coder
```
### Chat Template
```python
from transformers import AutoTokenizer
model_name = 'nvidia/Nemotron-Cascade-2-30B-A3B'
tokenizer = AutoTokenizer.from_pretrained(model_name)
'''
single-turn example
'''
messages = [
{"role": "system", "content": "You are a helpful and harmless assistant.\n\nYou are not allowed to use any tools"},
{"role": "user", "content": "calculate 1+1?"}
]
# thinking mode
prompt_thinking = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)
# prompt_thinking = '<|im_start|>system\nYou are a helpful and harmless assistant.\n\nYou are not allowed to use any tools<|im_end|>\n<|im_start|>user\ncalculate 1+1?<|im_end|>\n<|im_start|>assistant\n\n'
# instruct mode
prompt_instruct = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)
# prompt_instruct = '<|im_start|>system\nYou are a helpful and harmless assistant.\n\nYou are not allowed to use any tools<|im_end|>\n<|im_start|>user\ncalculate 1+1?<|im_end|>\n<|im_start|>assistant\n'
'''
multi-turn example
'''
messages = [
{"role": "system", "content": "You are a helpful and harmless assistant.\n\nYou are not allowed to use any tools"},
{"role": "user", "content": "calculate 1+1?"},
{"role": "assistant", "content": "THINKING_CONTENT\nTo calculate \\(1 + 1\\):\n\n1. **Identify the operation**: This is a basic addition problem involving two integers.\n2. **Perform the addition**: \n \\(1 + 1 = 2\\).\n\n**Result**: \\(\\boxed{2}\\)",},
{"role": "user", "content": "what about 2+2"}
]
# thinking mode
prompt_thinking = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)
# prompt_thinking = '<|im_start|>system\nYou are a helpful and harmless assistant.\n\nYou are not allowed to use any tools<|im_end|>\n<|im_start|>user\ncalculate 1+1?<|im_end|>\n<|im_start|>assistant\n\nTo calculate \\(1 + 1\\):\n\n1. **Identify the operation**: This is a basic addition problem involving two integers.\n2. **Perform the addition**: \n \\(1 + 1 = 2\\).\n\n**Result**: \\(\\boxed{2}\\)<|im_end|>\n<|im_start|>user\nwhat about 2+2<|im_end|>\n<|im_start|>assistant\n\n'
# instruct mode
prompt_instruct = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)
# prompt_instruct = '<|im_start|>system\nYou are a helpful and harmless assistant.\n\nYou are not allowed to use any tools<|im_end|>\n<|im_start|>user\ncalculate 1+1?<|im_end|>\n<|im_start|>assistant\n\nTo calculate \\(1 + 1\\):\n\n1. **Identify the operation**: This is a basic addition problem involving two integers.\n2. **Perform the addition**: \n \\(1 + 1 = 2\\).\n\n**Result**: \\(\\boxed{2}\\)<|im_end|>\n<|im_start|>user\nwhat about 2+2<|im_end|>\n<|im_start|>assistant\n'
```
### Python Tool Use
```python
model_name = 'nvidia/Nemotron-Cascade-2-30B-A3B'
tokenizer = AutoTokenizer.from_pretrained(model_name)
SYSTEM_PROMPT = """# Tools
You have access to the following functions:
stateful_python_code_exec
Call this function to execute Python code in a stateful Jupyter notebook environment. Python will respond with the output of the execution or time out after 120.0 seconds.
code
string
Code to execute
["code"]
If you choose to call a function ONLY reply in the following format with NO suffix:
value_1
This is the value for the second parameter
that can span
multiple lines
Reminder:
- Function calls MUST follow the specified format: an inner block must be nested within XML tags
- Required parameters MUST be specified
- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after
- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls
"""
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": "Solve the following math problem. Put your answer inside \\boxed{}.\n\nIn a school with 2008 students, each student is a member of certain committees. Each committee has at most 1004 members, and every two students are in at least one common committee. Determine the smallest possible number of committees in the school."}
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)
print(prompt)
```
### Agentic Usage
```python
model_name = 'nvidia/Nemotron-Cascade-2-30B-A3B'
tokenizer = AutoTokenizer.from_pretrained(model_name)
SYSTEM_PROMPT = """You are a customer service agent that helps the user. The policy that determines how you should respond to requests from users is described below between the and tags.
In each turn you can either:
- Send a message to the user.
- Make a tool call.
You cannot do both at the same time.
_NEED_TO_ADD_POLICY_HERE_
Try to be helpful and always follow the policy.
# Tools
You have access to the following functions:
_NEED_TO_ADD_FUNCTION_NAME_1_
_FUNCTION_DESCRIPTION_
_NEED_TO_ADD_PARAMETER_NAME_1_
_PARAMETER_TYPE_
_PARAMETER_DESCRIPTION_
_PARAMETER_TITLE_
_NEED_TO_ADD_PARAMETER_NAME_2_
_PARAMETER_TYPE_
_PARAMETER_DESCRIPTION_
_PARAMETER_TITLE_
...... (_MORE_PARAMETERS_TO_ADD_)
...... (_MORE_FUNCTIONS_TO_ADD_)
"""
messages = [
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": "Hello, I'm calling regarding my upcoming stay at your hotel. My guest ID is G90920 and booking ID is B11246 for a Deluxe room on June 5th. I'm traveling with three 6-month-old triplets and need to request three infant cribs for our room. It's currently 30 hours before check-in—could you please confirm if this is feasible and if there are quiet room options available for families with infants?"}
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)
print(prompt)
```
## Release Date
Mar 19, 2026
## License
Your use of this model is governed by the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/).
## Citation
```
@article{Nemotron_Cascade_2,
title={Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation},
author={Yang, Zhuolin and Liu, Zihan and Chen, Yang and Dai, Wenliang and Wang, Boxin and Lin, Sheng-Chieh and Lee, Chankyu and Chen, Yangyi and Jiang, Dongfu and He, Jiafan and Pi, Renjie and Lam, Grace and Lee, Nayeon and Bukharin, Alexander and Shoeybi, Mohammad and Catanzaro, Bryan and Ping, Wei},
year={2026}
}
```