Nemotron-3.5-Bolt-E4B-BF16

Model Overview

Model Developer: NVIDIA Corporation

Model Dates: December 2025 - May 2026

Data Freshness:

• The post-training data has a cutoff date of May 2026.

Description

Nemotron-3.5-Bolt-E4B-BF16 is a large language model (LLM) further post-trained from Gemma-4-E4B-it by NVIDIA. This is the full-precision (BF16) release of Nemotron 3.5 Bolt E4B.

This model is ready for commercial or non‑commercial use.

License/Terms of Use

GOVERNING HOSTING TERMS: Use of the API is governed by the NVIDIA API Trial Terms of Service. Use of the model is governed by the OpenMDW License Agreement, version 1.1.

ADDITIONAL INFORMATION: Apache License, Version 2.0.

Benchmarks

Reasoning Benchmark Evaluations

We evaluated our model on the following benchmarks:

Accuracy numbers are measured by NVIDIA under a consistent harness (NeMo Gym / Nemo Evaluator SDK); they may differ from vendors' self-reported numbers.

Deployment Geography: Global

Use Case

Nemotron-3.5-Bolt-E4B-BF16 is the full-precision release of a general purpose reasoning and chat model, and is intended primarily for customization and post-training rather than direct production inference. It is intended to be used by developers who want to: post-train the model on their own data (SFT, RL via NeMo RL and NeMo Gym, or distillation), adapt it to a domain or task, produce quantized variants (NVFP4, W4A16, GGUF) for their own deployment targets, or run full-precision research and evaluation. English and coding languages are the primary languages, with Spanish, French, German, Italian, and Japanese also supported.

Release Date

Build.NVIDIA.com: 10/23/2026 via Nemotron 3.5 Bolt E4B

Hugging Face 10/23/2026 via https://huggingface.co/nvidia/Nemotron-3.5-Bolt-E4B-BF16

Model Architecture

• Architecture Type: Gemma 4
• Network Architecture: Gemma-4-E4B
• Number of model parameters: 8B (4B effective parameters)

Model Design

The model was further post-trained from Gemma-4-E4B-it.

Training Methodology

Reinforcement Learning

• The model underwent multi-environment reinforcement learning using GRPO (Group Relative Policy Optimization) across instruction following, multi-step tool use, multi-turn conversations, and structured output environments. It utilized an asynchronous RL architecture that decouples training from inference.
• Software used for reinforcement learning: NeMo RL, NeMo Gym

Nemotron-3.5-Bolt-E4B-BF16 is a result of the above work.

Input

• Input Type(s): Text
• Input Format(s): String
• Input Parameters: One-Dimensional (1D): Sequences
• Other Properties Related to Input: Maximum context length for this WCE endpoint: up to 64K tokens. Supported languages include English, Spanish, French, German, Italian, and Japanese.

Output

• Output Type(s): Text
• Output Format: String
• Output Parameters: One-Dimensional (1D): Sequences
• Other Properties Related to Output: A text response, with an optional reasoning trace and tool calls, in OpenAI-compatible chat format.

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

• Runtime Engine(s): PyTorch
• Supported Hardware Microarchitecture Compatibility: NVIDIA Ampere - A100; NVIDIA Blackwell; NVIDIA Hopper - H100-80GB
• Supported 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)

• GA (10/23/2026)

Asynchronous Tool-Calling

• Transmit asynchronous functions via the standard OpenAI tools structure.
• Retrieve assistant function executions directly from message.tool_calls.
• Restrict generation to a single tool invocation per assistant turn.
• Assign a persistent task identifier to each decoupled job execution.
• Provide subsequent execution outputs as a user role turn prefixed with something like "TOOL RESULT:".
• Avoid submitting secondary tool turns for requests with an active pending response.
• Maintain tracked status sets for unresolved and finalized job identifiers.
• Synthesize the final response only after all background processes finish and zero calls remain outstanding.

Example:

Assistant tool-call

{
  "role": "assistant",
  "tool_calls": [{
    "id": "call_1",
    "type": "function",
    "function": {
      "name": "build_report",
      "arguments": "{\"_task_id\":\"task_1\",\"scope\":\"weekly\"}"
    }
  }]
}

Immediate acknowledgement with pending result

{
  "role": "tool",
  "tool_call_id": "call_1",
  "content": "{\"status\":\"pending\",\"task_id\":\"task_1\",\"function\":\"build_report\"}"
}

Later tool result completion result

{
  "role": "user",
  "content": "ASYNC TOOL UPDATE: {\"status\":\"completed\",\"task_id\":\"task_1\",\"function\":\"build_report\",\"response\":\"Report is ready.\"}"
}

The model processes the finalized payload to resume generation.

Prompting & Usage Guidelines

General guidelines for using Gemma-Nemotron in conversational and agentic applications.

Prompt Structure & Content

• Store persistent behavior constraints and operational guidelines in system prompts.
• Use explicit headers to organize roles, tasks, tool configs, and formatting safety rules.
• Keep the primary objective clear and concise.
• Write unambiguous, verifiable instructions and resolve conflicting guidance.
• Remove redundant guidance and repetitive few-shot examples.
• Use explicit constraints (e.g., "limit responses to 1-2 sentences") instead of vague requests (e.g., "be brief").
• Avoid synthetic or non-model outputs in "assistant" turns.
• Adhere strictly to standard system/user/assistant/tool turn sequences.

Conveying Environment State in Stateful Applications

• Prefix context blocks with explicit structural markers (e.g., "# CURRENT STATE").
• Clearly specify what part of the context belongs to the model and what part refers to information for other entities like the user.
• When state updates end a message, restate the core goal alongside them.

Instructing the Model

• Define required behaviors with precise language and mandate necessary tool calls.
• Provide fallback protocols for requests exceeding available tool capabilities.
• Separate output formatting rules from task directives, and use the official chat template.

Conversational Responses

• Set conversational_mode=True for concise, low-formatting outputs suited for audio or spoken interaction.
• Set or leave the flag to False to retain standard generation style.

Training, Testing, and Evaluation Datasets

Training

Data Modality: Text Training Data Size: More than 20 Trillion Tokens Dataset partition: Training [100%], testing [0%], validation [0%] Time period for training data collection: 2013 to December 2025 Time period for testing data collection: 2013 to December 2025 Time period for validation data collection: 2013 to December 2025 Data Collection Method by dataset: Hybrid: Automated, Manually-Collected, Synthetic Labeling Method by dataset: Hybrid: Automated, Manually-Labeled, Synthetic

The post-training corpus for Nemotron-3.5-Bolt-E4B-BF16 consists of high-quality curated and synthetically-generated data. Primary languages used for post-training include English, French, German, Italian, Japanese, Spanish, and Chinese.

These datasets, such as FinePDFs, EssentialWeb, HotpotQA, SQuAD, and HelpSteer3, do not collectively or exhaustively represent all demographic groups (and proportionally therein). For instance, these datasets do not contain explicit mentions of demographic classes such as age, gender, or ethnicity in 64-99% of samples, depending on the source. In the subset where such terms are present, document-based datasets (FinePDFs and EssentialWeb) contain representational skews, such as references to "male" outnumbering those to "female", and mentions of "White" as the most frequent among ethnic identifiers (comprising 43-44% of ethnicity mentions). To mitigate these imbalances, we recommend considering evaluation techniques such as bias audits, fine-tuning with demographically balanced datasets, and mitigation strategies like counterfactual data augmentation to align with the desired model behavior. This evaluation used a 3,000-sample subset per dataset, identified as the optimal threshold for maximizing embedder accuracy.

During post-training, we generate synthetic data by distilling trajectories, solutions, and translations from strong teacher models and agent systems, often grounded in real tasks or documents and aggressively filtered for quality. For math, code, and science, we start from curated problem sets and use open source permissive models such as GPT-OSS-120B to produce step-by-step reasoning traces, candidate solutions, best-of-n selection traces, and verified CUDA kernels. For long-context and science, we build synthetic QA and reasoning data by retrieving passages from long documents, generating MCQ/OpenQA questions and answers, and paraphrasing them into multiple prompt/response formats to ensure diversity. Across all pipelines we stack automated verification—compilers, numerical checks, language identification—to ensure our data is high quality.

For all domains, we apply a unified data filtering pipeline to ensure that only high-quality, license-compliant, and verifiable samples are used for post-training. We first discard malformed examples using structural checks (e.g., missing tool definitions when tool calls are present). We then aggressively filter reasoning traces exhibiting pathological repetition, such as repeated n-grams within a sliding window or across the entire trajectory, which we found to be a strong indicator of malformed or low-quality reasoning. Finally, based on internal audits of synthetically generated datasets, we observed that some teacher models occasionally produce reasoning traces and final responses that implicitly align with specific political entities or promote nationalistic narratives. To mitigate this, we apply targeted keyword- and regex-based filters and remove all trajectories matching such behavior.

Alongside the model, we release our final pre-training and post-training data, as outlined in this section. For ease of analysis, there is a sample set that is ungated. For all remaining code, math and multilingual data, gating and approval is required, and the dataset is permissively licensed for model training purposes.

Testing Datasets:

Data Collection Method by dataset

• Hybrid: Automated, Manually-Collected, Synthetic Labeling Method by dataset
  
• Hybrid: Automated, Manually-Labeled, Synthetic Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI. These benchmarks test model capabilities on tasks such as tool-calling and instruction following.

Evaluation Datasets:

Data Collection Method by dataset

• Hybrid: Automated, Manually-Collected, Synthetic Labeling Method by dataset
  
• Hybrid: Automated, Manually-Labeled, Synthetic Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI. These benchmarks test model capabilities on tasks such as tool-calling and instruction following.

Inference

Acceleration Engine: vLLM with NVIDIA Dynamo (Preview API).

Test Hardware: Reported for the base model:

• NVIDIA Hopper
  • 1-8x H100
  • 1-8x H200
• NVIDIA Blackwell
  • GB200
  • GeForce RTX 5090

Preview API Deployment Hardware: NVIDIA L40S.

Recommended sampling settings are Temperature 1.0 and Top_P 0.95 in both thinking enabled and disabled modes.

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. 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.

Users are responsible for model inputs and outputs. Users are responsible for ensuring safe integration of this model, including implementing guardrails as well as other safety mechanisms, prior to deployment.

We advise against circumvention of any provided safety guardrails contained in the Model without a substantially similar guardrail appropriate for your use case. For more details: Safety and Explainability Subcards.

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.