dusersad12 / NovaMind-1-Release

NovaMind-1

1. Introduction

NovaMind-1 is Nova Labs' newest assistant model, and this release is a big step up from NovaMind-0. We spent considerably more compute on post-training this time and reworked the optimization recipe, which shows up most clearly in reasoning-heavy workloads: math, code, and multi-step logic. On our internal suites it now sits comfortably in the same band as several of the strongest open models available today.

The headline story of this generation is depth of thought. On the AIME 2025 problem set we measure 87.5% accuracy for NovaMind-1 against 70% for NovaMind-0, and the average reasoning budget grew from roughly 12K tokens per question to 23K. In plain terms: the model is willing to think for much longer before it commits to an answer, and that patience pays off.

Two more things we worked on explicitly for this release: the hallucination rate on factual prompts is measurably lower than NovaMind-0, and function/tool calling is now a first-class capability rather than a bolt-on.

2. Evaluation Results

Comprehensive Benchmark Results

Overall Performance Summary

Across every category we track, NovaMind-1 clears NovaMind-0's numbers, with the widest margins concentrated in the reasoning and generation buckets.

3. Chat Website & API Platform

You can try NovaMind-1 through the chat interface and the hosted API on the Nova Labs website; both expose the same weights as this release.

4. How to Run Locally

NovaMind-1 loads with the standard transformers tooling - no custom modeling code is required.

Usage notes for this generation:

1. A system prompt is now supported and recommended.
2. You no longer need to prepend special tokens to steer the model into thinking mode; that workaround belongs to NovaMind-0.

The Small variant shares the architecture of its base model and reuses the main tokenizer, so everything below applies to it unchanged.

System Prompt

We recommend starting every session with a system prompt that anchors the date:

You are NovaMind-1, a helpful AI assistant.
Today is {current date}.

For example:

You are NovaMind-1, a helpful AI assistant.
Today is May 28, 2025, Monday.

Temperature

A sampling temperature of $T_{model} = 0.6$ works best for most tasks.

Prompt Templates for File Uploads and Web Search

When the user attaches a file, wrap the content like this, with {file_name} , {file_content} and {question} as the arguments:

file_template = \
"""[file name]: {file_name}
[file content begin]
{file_content}
[file content end]
{question}"""

When the answer should lean on web search results, use the following template, with {search_results} , {cur_date} , and {question} as the arguments:

search_answer_en_template = \
'''# The following contents are the search results related to the user's message:
{search_results}
In the search results I provide to you, each result is formatted as [webpage X begin]...[webpage X end], where X represents the numerical index of each article. Please cite the context at the end of the relevant sentence when appropriate. Use the citation format [citation:X] in the corresponding part of your answer. If a sentence is derived from multiple contexts, list all relevant citation numbers, such as [citation:3][citation:5]. Be sure not to cluster all citations at the end; instead, include them in the corresponding parts of the answer.
When responding, please keep the following points in mind:
- Today is {cur_date}.
- Not all content in the search results is closely related to the user's question. You need to evaluate and filter the search results based on the question.
- For listing-type questions (e.g., listing all flight information), try to limit the answer to 10 key points and inform the user that they can refer to the search sources for complete information. Prioritize providing the most complete and relevant items in the list. Avoid mentioning content not provided in the search results unless necessary.
- For creative tasks (e.g., writing an essay), ensure that references are cited within the body of the text, such as [citation:3][citation:5], rather than only at the end of the text. You need to interpret and summarize the user's requirements, choose an appropriate format, fully utilize the search results, extract key information, and generate an answer that is insightful, creative, and professional. Extend the length of your response as much as possible, addressing each point in detail and from multiple perspectives, ensuring the content is rich and thorough.
- If the response is lengthy, structure it well and summarize it in paragraphs. If a point-by-point format is needed, try to limit the answer to 5 points and merge related points.
- For objective Q&A, if the answer is very brief, you may add one or two related sentences to enrich the content.
- Choose an appropriate and visually appealing format for your response based on the user's requirements and the content of the answer, ensuring strong readability.
- Your answer should synthesize information from multiple relevant webpages and avoid repeatedly citing the same webpage.
- Unless the user requests otherwise, your response should be in the same language as the user's question.
# The user's message is:
{question}'''

5. License

Both the code and the NovaMind-1 weights are released under the MIT License . Commercial use and distillation are permitted.

6. Contact

Questions or bug reports? Open an issue in the Nova Labs repository or write to [email protected] .