_PARAMETER_TITLE

[{"_1":2,"_179":-5,"_180":-5},"loaderData",{"_3":4,"_17":18,"_25":26,"_156":157},"root",{"_5":-5,"_6":-7,"_7":8,"_9":-7,"_10":11,"_12":-7,"_13":14,"_15":16},"currentUser","primaryEmail","siteVerification","6LeIxAcTAAAAAJcZVRqyHh71UMIEGNQ_MXjiZKhI","toast","enableDev",false,"incidentStatus","theme","system","contrast","normal","routes/_layout",{"_19":-5,"_20":21,"_22":21,"_23":11,"_24":21},"username","isAskGordonEnabled",true,"isUsingNewTemplate","showMinimalLanding","showHoneypotTrap","routes/_layout.r.$namespace.$reponame",{"_27":28,"_29":-5,"_30":11,"_31":11,"_32":33,"_34":35,"_36":-5,"_37":11,"_38":39,"_90":91,"_114":115,"_145":-5,"_146":11,"_147":21,"_148":11,"_149":-5,"_150":11,"_151":-5,"_152":153,"_154":116,"_155":-7},"canonicalUrl","https://hub.docker.com/r/ai/nemotron-cascade-2/tags","mcpServer","canSeeDockerfile","canSeeAutobuilds","namespace","ai","reponame","nemotron-cascade-2","cloudBuildSource","isAutobuildsAvailable","repo",{"_40":41,"_42":43,"_45":46,"_47":48,"_49":11,"_50":51,"_52":53,"_54":55,"_57":35,"_32":33,"_58":59,"_60":61,"_62":63,"_64":65,"_66":67,"_68":-5,"_69":63,"_70":71,"_72":11,"_73":74,"_75":76,"_81":11,"_82":83,"_87":33,"_88":89},"categories",[],"content_types",[44],"unrecognized","date_registered","2026-08-04T16:16:37.577324Z","description","","is_private","last_updated","2026-08-29T19:23:46.289669Z","last_modified","2026-10-02T16:02:35.760215Z","media_types",[56],"application/vnd.cncf.model.manifest.v1+json","name","pull_count",8180,"repository_type","image","star_count",0,"status",1,"status_description","active","source","collaborator_count","full_description","\n# Nemotron-Cascade-2-30B-A3B\n\n

\n\n[![Technical Report](https://img.shields.io/badge/2603.19220-Technical_Report-blue)](https://arxiv.org/abs/2603.19220)\n[![SFT Dataset](https://img.shields.io/badge/🤗-SFT_Datset-blue)](https://huggingface.co/datasets/nvidia/Nemotron-Cascade-2-SFT-Data)\n[![RL Dataset](https://img.shields.io/badge/🤗-RL_Datset-blue)](https://huggingface.co/datasets/nvidia/Nemotron-Cascade-2-RL-data)\n[![Models](https://img.shields.io/badge/🤗-Models-blue)](https://huggingface.co/collections/nvidia/nemotron-cascade-2)\n

\n\n $\"main_fig\"$ \n\n\n## Introduction\nWe'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.\n\n\n## Benchmark Results\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

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

\n\n

\n* † Numbers in brackets refers to Tool-Integrated Reasoning (TIR) results.
\n* ‡ For the baseline models, we evaluate them using the recommended settings when official numbers are unavailable; otherwise, we report the official numbers directly.
\n

\n\n\n## Quick Start\n\n- 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. \n\n- Nemotron-Cascade-2-30B-A3B supports up to a 1M-token context length.\n\n- Nemotron-Cascade-2-30B-A3B does not currently support OpenCode; it primarily supports OpenHands for agentic coding and SWE tasks.\n\n- 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.\n\n- 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 ``.\n\n- We recommend setting the sampling parameters to temperature = 1.0 and top_p = 0.95.\n\n\n### vLLM setup\n\nRequires vLLM version >= 0.17.1. The following will create API endpoints at `http://localhost:8000/v1`:\n\n- **Standard version**: Use the following command to create an API endpoint with a maximum context length of 1M tokens.\n\n ```shell\n 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 nano_v3 --mamba-ssm-cache-dtype float32 --port 8000 --trust_remote_code\n ```\n\n- **Tool Call**: Use the following command to enable tool support.\n\n ```shell\n 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 nano_v3 --mamba-ssm-cache-dtype float32 --port 8000 --trust_remote_code --enable-auto-tool-choice --tool-call-parser qwen3_coder\n ```\n\n**UPDATE (July 9, 2026)**: We changed `--reasoning-parser` from `nemotron_v3` to `nano_v3`. The `nano_v3` parser simply splits the output on ``: content before `` is treated as `reasoning content`, and content after it is treated as `summary content`. If `` is not present, the entire output remains in `reasoning content`. In contrast, the `nemotron_v3` parser treats `` or `` as an implicit end-of-reasoning boundary when `` is missing. This difference affects tool-calling scenarios.\n\n\n### Chat Template\n\n```python\nfrom transformers import AutoTokenizer\n\nmodel_name = 'nvidia/Nemotron-Cascade-2-30B-A3B'\ntokenizer = AutoTokenizer.from_pretrained(model_name)\n\n'''\nsingle-turn example\n'''\nmessages = [\n {\"role\": \"system\", \"content\": \"You are a helpful and harmless assistant.\\n\\nYou are not allowed to use any tools\"},\n {\"role\": \"user\", \"content\": \"calculate 1+1?\"}\n]\n\n# thinking mode\nprompt_thinking = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)\n# 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'\n\n# instruct mode\nprompt_instruct = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)\n# 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'\n\n'''\nmulti-turn example\n'''\nmessages = [\n {\"role\": \"system\", \"content\": \"You are a helpful and harmless assistant.\\n\\nYou are not allowed to use any tools\"},\n {\"role\": \"user\", \"content\": \"calculate 1+1?\"},\n {\"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}\\\$\",},\n {\"role\": \"user\", \"content\": \"what about 2+2\"}\n]\n\n# thinking mode\nprompt_thinking = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)\n# 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'\n\n# instruct mode\nprompt_instruct = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)\n# 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'\n```\n\n### Python Tool Use\n\n```python\nmodel_name = 'nvidia/Nemotron-Cascade-2-30B-A3B'\ntokenizer = AutoTokenizer.from_pretrained(model_name)\n\nSYSTEM_PROMPT = \"\"\"# Tools\n\nYou have access to the following functions:\n\n\n\nstateful_python_code_exec\nCall 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.\n\n\ncode\nstring\nCode to execute\n\n[\"code\"]\n\n\n\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n\n\n\nvalue_1\n\n\nThis is the value for the second parameter\nthat can span\nmultiple lines\n\n\n\n\n\nReminder:\n- Function calls MUST follow the specified format: an inner block must be nested within XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- 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\n\"\"\"\n\nmessages = [\n {\"role\": \"system\", \"content\": SYSTEM_PROMPT},\n {\"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.\"}\n]\n\nprompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)\nprint(prompt)\n```\n\n### Agentic Usage\n\n```python\nmodel_name = 'nvidia/Nemotron-Cascade-2-30B-A3B'\ntokenizer = AutoTokenizer.from_pretrained(model_name)\n\nSYSTEM_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.\n\nIn each turn you can either:\n- Send a message to the user.\n- Make a tool call.\nYou cannot do both at the same time.\n\n\n_NEED_TO_ADD_POLICY_HERE_\n\n\nTry to be helpful and always follow the policy.\n\n# Tools\n\nYou have access to the following functions:\n\n\n\n_NEED_TO_ADD_FUNCTION_NAME_1_\n_FUNCTION_DESCRIPTION_\n\n\n_NEED_TO_ADD_PARAMETER_NAME_1_\n_PARAMETER_TYPE_\n_PARAMETER_DESCRIPTION_\n_PARAMETER_TITLE_\n\n\n_NEED_TO_ADD_PARAMETER_NAME_2_\n_PARAMETER_TYPE_\n_PARAMETER_DESCRIPTION_\n_PARAMETER_TITLE_\n\n...... (_MORE_PARAMETERS_TO_ADD_)\n\n\n...... (_MORE_FUNCTIONS_TO_ADD_)\n\n\"\"\"\n\nmessages = [\n {\"role\": \"system\", \"content\": SYSTEM_PROMPT},\n {\"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?\"}\n]\n\nprompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=True)\nprint(prompt)\n```\n\n## Release Date\nMar 19, 2026\n\n\n## License\nYour 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/).\n\n\n## Citation\n```\n@article{Nemotron_Cascade_2,\n title={Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation},\n 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},\n year={2026}\n}\n```","has_starred","hub_user","ericcurtin044","immutable_tags_settings",{"_77":11,"_78":79},"enabled","rules",[80],".*","is_automated","permissions",{"_84":21,"_85":11,"_86":11},"read","write","admin","user","storage_size",63174603051,"orgData",{"_92":93,"_94":95,"_96":97,"_98":33,"_99":48,"_100":101,"_102":103,"_104":105,"_106":107,"_108":109,"_110":48,"_111":112,"_113":21},"id","24d9c867e1a94da6b95aea2650d8ac17","uuid","24d9c867-e1a9-4da6-b95a-ea2650d8ac17","type","organization","orgname","full_name","location","San Francisco, CA, USA","company","Docker","profile_url","https://www.docker.com","date_joined","2024-10-01T12:49:08.457388Z","gravatar_url","https://www.gravatar.com/avatar/56b6e15d486f6407690a6cb56e3fa68e?s=80&r=g&d=mm","gravatar_email","badge","verified_publisher","is_active","recentTags",[116,140],{"_117":44,"_118":119,"_120":121,"_122":123,"_92":124,"_125":126,"_50":127,"_128":119,"_129":74,"_130":56,"_57":131,"_132":133,"_134":135,"_136":127,"_137":138,"_139":21},"content_type","creator",36638321,"digest","sha256:d48d35f2aaf99799401b5ac65b7067dfaadc2335d9db407ff17bf7a9851731da","full_size",63174589639,1259118589,"images",[],"2026-08-29T18:05:50.998971Z","last_updater","last_updater_username","media_type","latest","repository",32025223,"tag_last_pulled","2026-10-02T16:05:03.543741587Z","tag_last_pushed","tag_status","inactive","v2",{"_117":44,"_118":119,"_120":121,"_122":123,"_92":141,"_125":142,"_50":143,"_128":119,"_129":74,"_130":56,"_57":144,"_132":133,"_134":135,"_136":143,"_137":138,"_139":21},1259118682,[],"2026-08-29T19:23:46.167893Z","safetensors","sbxKitMetadata","isPotentialSpam","isHubSbxKitsEnabled","showSbxLogoOverride","sbxTrademarkText","displaySecurityMetadata","tagAttestations","tagsHaveAttestations",[],"selectedTag","attestationsUrl","routes/_layout.r.$namespace.$reponame.tags",{"_38":158,"_165":166,"_152":177,"_178":11},{"_40":159,"_42":160,"_45":46,"_47":48,"_49":11,"_50":51,"_52":53,"_54":161,"_57":35,"_32":33,"_58":59,"_60":61,"_62":63,"_64":65,"_66":67,"_68":-5,"_69":63,"_70":71,"_72":11,"_73":74,"_75":162,"_81":11,"_82":164,"_87":33,"_88":89},[],[44],[56],{"_77":11,"_78":163},[80],{"_84":21,"_85":11,"_86":11},"tags",{"_167":168,"_169":-5,"_170":-5,"_171":172},"count",2,"next","previous","results",[173,175],{"_117":44,"_118":119,"_120":121,"_122":123,"_92":141,"_125":174,"_50":143,"_128":119,"_129":74,"_130":56,"_57":144,"_132":133,"_134":135,"_136":143,"_137":138,"_139":21},[],{"_117":44,"_118":119,"_120":121,"_122":123,"_92":124,"_125":176,"_50":127,"_128":119,"_129":74,"_130":56,"_57":131,"_132":133,"_134":135,"_136":127,"_137":138,"_139":21},[],[],"canDeleteTags","actionData","errors"]

ai/nemotron-cascade-2

Tags for ai/nemotron-cascade-2