NVIDIALanguage model
Nemotron Mini 4B Instruct
Nemotron Mini 4B Instruct has no overall index value yet because there are too few independent benchmark results.
PopularityRank 452 of 785
33out of 100-2 vs. last month
based on search, usage and attention
Summary
Nemotron Mini 4B Instruct comes from NVIDIA’s pruning, distillation, and quantization of Minitron 4B, which itself was compressed from Nemotron 4 15B.
The model card recommends a specific chat template and lists a 4,096-token context for the released checkpoint.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 26 published benchmark results
Page 1 of 3
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
GSM8K-Hi (Show measurement, test conditions, and source)
- Source value
- 32.22
- Score
- 32.22
- Metric
- accuracy on the paper's 1-100 scale
- Unit
- %
- Benchmark version
- BHASHA 2025 Table 2; custom Hindi adaptation; LM-Eval-Harness; 1,319 samples
- Direction
- Higher is better
- Sample
- 1319
- Source type
- vendor-reported
- Evaluator
- Kamath et al., Benchmarking Hindi LLMs, BHASHA 2025 (NVIDIA authors)
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Nemotron-Mini-4B-Instruct row. The paper says GSM8K-Hi uses LM-Eval-Harness and reports accuracy on a 1-100 scale; shot count, model revision, and decoding settings are not stated.
- Context
- These are results on new Hindi-specific benchmark adaptations, not the original English benchmark sets. The paper does not pin the evaluated model revision or full inference setup.
IFEval-Hi (Show measurement, test conditions, and source)
- Source value
- 36.08
- Score
- 36.08
- Metric
- accuracy on the paper's 1-100 scale
- Unit
- %
- Benchmark version
- BHASHA 2025 Table 2; custom Hindi adaptation; 848 samples
- Direction
- Higher is better
- Sample
- 848
- Source type
- vendor-reported
- Evaluator
- Kamath et al., Benchmarking Hindi LLMs, BHASHA 2025 (NVIDIA authors)
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Nemotron-Mini-4B-Instruct row in the custom Hindi IFEval-Hi suite. The paper defines the table score as accuracy on a 1-100 scale; model revision, prompt, and decoding settings are not stated.
- Context
- These are results on new Hindi-specific benchmark adaptations, not the original English benchmark sets. The paper does not pin the evaluated model revision or full inference setup.
Mobile-MMLU (Show measurement, test conditions, and source)
- Source value
- 35.1
- Score
- 35.1
- Metric
- accuracy
- Unit
- %
- Benchmark version
- Mobile-MMLU paper, Table 2; zero-shot
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Mobile-MMLU paper, Proceedings of Machine Learning Research, Vol. 03-1
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper reports zero-shot results in Table 2 and identifies accuracy as its primary metric. The row names the exact Nemotron-Mini-4B-Instruct checkpoint. A table-specific sample count, immutable weight revision, and full decoding parameters are not reported.
- Context
- The source does not publish the number of evaluated items for these tables or an exact immutable model revision. No confidence interval is reported.
Mobile-MMLU-Pro (Show measurement, test conditions, and source)
- Source value
- 30.8
- Score
- 30.8
- Metric
- accuracy
- Unit
- %
- Benchmark version
- Mobile-MMLU paper, Table 2; zero-shot
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Mobile-MMLU paper, Proceedings of Machine Learning Research, Vol. 03-1
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. The paper reports zero-shot results in Table 2 and identifies accuracy as its primary metric. The row names the exact Nemotron-Mini-4B-Instruct checkpoint. A table-specific sample count, immutable weight revision, and full decoding parameters are not reported.
- Context
- The source does not publish the number of evaluated items for these tables or an exact immutable model revision. No confidence interval is reported.
Mobile-MMLU OSQ (Show measurement, test conditions, and source)
- Source value
- 13.7
- Score
- 13.7
- Metric
- accuracy
- Unit
- %
- Benchmark version
- Mobile-MMLU paper, Table 3; OSQ; zero-shot
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Mobile-MMLU paper, Proceedings of Machine Learning Research, Vol. 03-1
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. Table 3 reports zero-shot Open-Style Questions (OSQ), answered directly without multiple-choice options. The paper evaluates OSQ answers with an LLM judge that returns Correct or Incorrect. It does not identify the exact judge revision, table-specific sample count, model weight revision, or full decoding parameters.
- Context
- The source does not publish the number of evaluated items for these tables or an exact immutable model revision. No confidence interval is reported.
Mobile-MMLU-Pro OSQ (Show measurement, test conditions, and source)
- Source value
- 10.7
- Score
- 10.7
- Metric
- accuracy
- Unit
- %
- Benchmark version
- Mobile-MMLU paper, Table 3; OSQ; zero-shot
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Mobile-MMLU paper, Proceedings of Machine Learning Research, Vol. 03-1
- Status
- active
- Retrieved at
- 2026-10-19
- Methodology
- Independent evaluation. Table 3 reports zero-shot Open-Style Questions (OSQ), answered directly without multiple-choice options. The paper evaluates OSQ answers with an LLM judge that returns Correct or Incorrect. It does not identify the exact judge revision, table-specific sample count, model weight revision, or full decoding parameters.
- Context
- The source does not publish the number of evaluated items for these tables or an exact immutable model revision. No confidence interval is reported.
MT-Bench-Hi (Show measurement, test conditions, and source)
- Source value
- 3.44
- Score
- 3.44
- Metric
- LLM-as-a-judge rating on the paper's 1-10 scale
- Unit
- rating
- Benchmark version
- BHASHA 2025 Table 2; custom Hindi adaptation; 200 samples
- Direction
- Higher is better
- Sample
- 200
- Source type
- vendor-reported
- Evaluator
- Kamath et al., Benchmarking Hindi LLMs, BHASHA 2025 (NVIDIA authors)
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Nemotron-Mini-4B-Instruct row in the custom Hindi evaluation suite. The paper specifies a 1-10 LLM-as-a-judge scale, but not an exact model revision, judge, prompt, or decoding settings for this run.
- Context
- These are results on new Hindi-specific benchmark adaptations, not the original English benchmark sets. The paper does not pin the evaluated model revision or full inference setup.
ChatRAG-Hi (Show measurement, test conditions, and source)
- Source value
- 27.32
- Score
- 27.32
- Metric
- F1 score; scale not specified
- Unit
- No unit provided
- Benchmark version
- BHASHA 2025 Table 2; custom Hindi adaptation; 5,948 samples
- Direction
- Higher is better
- Sample
- 5948
- Source type
- vendor-reported
- Evaluator
- Kamath et al., Benchmarking Hindi LLMs, BHASHA 2025 (NVIDIA authors)
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Exact Nemotron-Mini-4B-Instruct row in the custom Hindi ChatRAG-Hi suite. The paper identifies the reported metric as F1 score but does not state its scale or the model's prompt and decoding settings.
- Context
- These are results on new Hindi-specific benchmark adaptations, not the original English benchmark sets. The paper does not pin the evaluated model revision or full inference setup.
IndicSentiment (Show measurement, test conditions, and source)
- Source value
- 90.01
- Score
- 90.01
- Metric
- F1-NLU score; source scale not stated
- Unit
- No unit provided
- Benchmark version
- IndoNLP 2025, Table 3; Hindi task suite
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Adapting Multilingual LLMs to Low-Resource Languages using Continued Pre-training and Synthetic Corpus (IndoNLP 2025), Table 3
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Table 3 Instruct-model column Nemotron-Mini-4B; zero-shot per caption; no revision, precision, prompt, or decoding settings are reported.
IndicCopa (Show measurement, test conditions, and source)
- Source value
- 66.01
- Score
- 66.01
- Metric
- F1-NLU score; source scale not stated
- Unit
- No unit provided
- Benchmark version
- IndoNLP 2025, Table 3; Hindi task suite
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Adapting Multilingual LLMs to Low-Resource Languages using Continued Pre-training and Synthetic Corpus (IndoNLP 2025), Table 3
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Table 3 Instruct-model column Nemotron-Mini-4B; zero-shot per caption; no revision, precision, prompt, or decoding settings are reported.
IndicXNLI (Show measurement, test conditions, and source)
- Source value
- 39.25
- Score
- 39.25
- Metric
- F1-NLU score; source scale not stated
- Unit
- No unit provided
- Benchmark version
- IndoNLP 2025, Table 3; Hindi task suite
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Adapting Multilingual LLMs to Low-Resource Languages using Continued Pre-training and Synthetic Corpus (IndoNLP 2025), Table 3
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Table 3 Instruct-model column Nemotron-Mini-4B; zero-shot per caption; no revision, precision, prompt, or decoding settings are reported.
IndicXParaphrase (Show measurement, test conditions, and source)
- Source value
- 83.74
- Score
- 83.74
- Metric
- F1-NLU score; source scale not stated
- Unit
- No unit provided
- Benchmark version
- IndoNLP 2025, Table 3; Hindi task suite
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Adapting Multilingual LLMs to Low-Resource Languages using Continued Pre-training and Synthetic Corpus (IndoNLP 2025), Table 3
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Table 3 Instruct-model column Nemotron-Mini-4B; zero-shot per caption; no revision, precision, prompt, or decoding settings are reported.
Page 1 of 3
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
The 100 most popular models of this category, 0 to 100, higher is more popular
20 of 101 models
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 7, 2026
Nemotron Mini 4B InstructMixtral 8x22B 33Amazon Nova Pro 33Aya Expanse 32B 33MiMo V2.6 Pro UltraSpeed 33
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as Nemotron Mini 4B Instruct, chosen by model line, developer, size, price and release date.
- Nemotron Nano 2 9BNVIDIA · same developer
- Nemotron Nano 2 VL 12BNVIDIA · same developer
- Llama 3.3 Nemotron Super 49B v1NVIDIA · same developer
- Llama 3.1 Nemotron 70B InstructNVIDIA · same developer
- Nemotron 3 NanoNVIDIA · same developer
- Llama 3.1 Nemotron Ultra 253BNVIDIA · same developer
- Llama 3.3 Nemotron Super 49B v1.5NVIDIA · same developer
- Mistral NemotronNVIDIA · same developer
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
Frequently asked questions about Nemotron Mini 4B Instruct
How large is the context window of Nemotron Mini 4B Instruct?
4,096 tokens. That is how much text Nemotron Mini 4B Instruct reads at once, about 10 A4 pages.
Page count as a rule of thumb from Artificial Analysis
GLM-5.1200K
DeepSeek-V3 0324164K
GLM-4.5131K
GPT OSS 120B131K
DeepSeek-V3.2128K
DeepSeek-R1128K
Nemotron Mini 4B InstructRank 103 of 1034K
Can I run Nemotron Mini 4B Instruct locally?
Yes, the weights are open.
Stated by NVIDIA
- Open weights
- yes
- License
- NVIDIA Open Model License
- Parameters
- 4B
- active per token
- n/a
Sources and data date
Show all 13 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.