nvidia/gpt-oss-120b-Eagle3-v3

Model Overview

Description:

The NVIDIA gpt-oss-120b is an Eagle-head variant of OpenAI’s gpt-oss-120b model, an autoregressive language model that uses a mixture-of-experts (MoE) architecture with 5 billion activated parameters and 120 billion total parameters. For more information, please check here. The NVIDIA gpt-oss-120b Eagle3 model incorporates Eagle speculative decoding with Model Optimizer.

This model is ready for commercial/non-commercial use.

License/Terms of Use:

Use of these model weights is governed by the nvidia-open-model-license. Additional Information: Apache License 2.0.

Deployment Geography:

Global

Use Case:

Developers designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications. Also suitable for typical instruction-following tasks.

This model improves the accuracy over all of gpt-oss-120b-Eagle3-short-context, gpt-oss-120b-Eagle3-long-context, and gpt-oss-120b-Eagle3-throughput. This model is recommended for all use-cases where one of the previous models is used.

Release Date:

Hugging Face May 8 2026 via https://huggingface.co/nvidia/gpt-oss-120b-Eagle3-v3

Model Architecture:

Architecture Type: Transformer
Network Architecture: gpt-oss-120b
Number of Model Parameters: 120 billion

Input:

Input Type(s): Text
Input Format(s): String
Input Parameters: One Dimensional (1D): Sequences
Other Properties Related to Input: None

Output:

Output Type(s): Text
Output Format: String
Output Parameters: One-Dimensional (1D): Sequences
Other Properties Related to Output: None

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration:

Supported Runtime Engine(s):

• TensorRT-LLM
  
• vLLM
  

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Blackwell
  

Preferred Operating System(s):

• Linux
  The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version(s):

This model is gpt-oss-120b Eagle3 version 3 quantized with Nvidia-ModelOpt v0.43.0

Training and Evaluation Datasets:

Training Dataset:

Links: Various datasets from NVIDIA's Nemotron Post-Training V3 Collection are used to train this model. Both the prompts and synthetically-generated responses in those datasets are used as-is. The following datasets are used:

• nvidia/Nemotron-Science-v1
• nvidia/Nemotron-Instruction-Following-Chat-v1
• nvidia/Nemotron-Competitive-Programming-v1
• nvidia/Nemotron-Math-v2
• nvidia/Nemotron-SFT-SWE-v2
• nvidia/Nemotron-SFT-Multilingual-v1
• nvidia/Nemotron-SFT-Agentic-v2
• nvidia/Nemotron-SFT-Instruction-Following-Chat-v2

Text Training Data Size: [1 Billion to 10 Trillion Tokens] Data Modality: Text Data Collection Method by Dataset: Hybrid: Synthetic, Human, Automated Labeling Method by Dataset: Hybrid: Synthetic, Human, Automated Properties: Total of ~2.9M samples, majority synthetic, others sourced from commercially-friendly datasets.

Evaluation Dataset:

Link: SPEED-Bench, for more details, see here Data Collection Method by Dataset: Hybrid: Human, Synthetic Labeling Method by Dataset: Hybrid: Human, Synthetic Properties: 880 curated multi-turn dialogue sequences, in 11 categories with 80 samples each.

The Eagle acceptance rate benchmark results (SPEED-Bench) with draft length 7 and temperature 0 are presented in the table below for medium reasoning:

Category	SPEED-Bench Acceptance Rate
coding	3.279
humanities	2.801
math	3.495
multilingual	3.387
qa	2.701
rag	3.085
reasoning	3.187
roleplay	2.306
stem	2.977
summarization	2.722
writing	2.516
Average	2.95

Eagle Speculative Decoding

Synthesized data was obtained from OpenAI's gpt-oss-120b model, which is then used to finetune the Eagle modules. This model is ready for inference with TensorRT-LLM and vLLM in Eagle speculative decoding mode. Eagle modules are used to predict candidate tokens beyond the next token. In the generation step, each forward Eagle module generates a distribution of tokens beyond the previous. Then, a tree-based attention mechanism samples some candidate sequences for the original model to validate. The longest accepted candidate sequence is selected so that more than 1 token is returned in the generation step. The number of tokens generated in each step is called acceptance length.

• The total size (in number of data points) 2.897M
  
• Total number of datasets 8
  
• Dataset partition: Training 100%
  

Inference:

Acceleration Engine: TensorRT-LLM v1.3.0rc11. vLLM v0.19.0 also supported.
Test Hardware: NVIDIA B200

Usage

TensorRT-LLM

To serve the checkpoint with TensorRT-LLM, follow the sample commands below with the TensorRT-LLM GitHub repo:

trtllm-serve <gpt-oss-120b checkpoint> --host 0.0.0.0 --port 8000 --backend pytorch --max_batch_size 32 --max_num_tokens 8192 --max_seq_len 8192 --tp_size 8 --extra_llm_api_options extra-llm-api-config.yml

extra-llm-api-config.yml is like this

enable_attention_dp: false
disable_overlap_scheduler: true
enable_autotuner: false

cuda_graph_config:
    max_batch_size: 1

speculative_config:
    decoding_type: Eagle
    max_draft_len: 7
    speculative_model_dir: nvidia/gpt-oss-120b-Eagle3-v3

kv_cache_config:
    enable_block_reuse: false

vLLM

To serve the checkpoint with vLLM, run the following:

vllm serve openai/gpt-oss-120b --speculative-config '{"method": "eagle3", "model": "nvidia/gpt-oss-120b-Eagle3-v3", "num_speculative_tokens": 7}'

Data Mixing

This model is trained in two phases. First, a short-context phase involving 2,697,247 total samples drawn from the listed datasets in various proportions. For each dataset the samples are drawn uniformly from the subset of samples with at most 4,096 total tokens.

Then a long-context phase involving 199,500 total samples is used with the same sampling strategy, but no limit on total sequence length.

The distribution of samples in each dataset/phase is pictured below:

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

For more detailed information on ethical considerations for this model, please see the Model Card++ Bias, Explainability, Safety & Security, and Privacy Subcards.

Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

SUBCARDS:

Explainability

Field:	Response:
Intended Task/Domain:	Text generation, reasoning, summarization, and question answering.
Model Type:	Text and Image-to-text transformer
Intended Users:	This model is intended for developers, researchers, and customers building/utilizing LLMs, while balancing accuracy and efficiency.
Output:	Text String(s)
Describe how the model works:	Generates text by predicting the next word or token based on the context provided in the input sequence using multiple self-attention layers
Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of:	Not Applicable
Technical Limitations & Mitigation:	The model was trained on data that contains toxic language and societal biases originally crawled from the internet. Therefore, the model may amplify those biases and return toxic responses especially when prompted with toxic prompts. Therefore, before deploying any applications of this model, developers should perform safety testing and tuning tailored to their specific applications of the model.
Verified to have met prescribed quality standards?	Yes
Performance Metrics:	Accuracy, Throughput, and user-side throughput
Potential Known Risk	The model may generate answers that may be inaccurate, omit key information, or include irrelevant or redundant text producing socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive.
Licensing:	Your usage is governed by the following license

Bias

Field:	Response:
Participation considerations from adversely impacted groups protected classes in model design and testing:	None
Measures taken to mitigate against unwanted bias:	None

Safety & Security

Field:	Response:
Model Application(s):	Chat, Instruction Following, Chatbot Development, Code Generation, Reasoning
Describe life critical application (if present):	Not Applicable
Use Case Restrictions:	Abide by the license
Model and Dataset Restrictions:	The Principle of least privilege (PoLP) is applied limiting access for dataset generation. Restrictions enforce dataset access during training, and dataset license constraints adhered to. Model checkpoints are made available on Hugging Face, and may become available on cloud providers' model catalog.

Privacy

Field:	Response:
Generatable or Reverse engineerable personal data?	No
Personal data used to create this model?	No
Was consent obtained for any personal data used?	Not Applicable
How often is dataset reviewed?	Before Release
Was data from user interactions with the AI model (e.g. user input and prompts) used to train the model?	No
Is there provenance for all datasets used in training?	Yes
Does data labeling (annotation, metadata) comply with privacy laws?	Yes
Is data compliant with data subject requests for data correction or removal, if such a request was made?	Not Applicable
Applicable NVIDIA Privacy Policy	https://www.nvidia.com/en-us/about-nvidia/privacy-policy/