Language modelOpen source
Nemotron Nano 2 9B
NVIDIA
- Released
- August 18, 2025
- Data date
- September 8, 2026
- Successor
- Nemotron 3 Nano
- Speed
- 143.1 tokens/sOutput throughput
- Price
- Input · Output / 1M tokens
- Input
- Text
- Output
- Text
The defining detail of Nemotron Nano 2 9B is its combination of Mamba2 and Transformer components in an open 9-billion-parameter model. Reasoning can be enabled, disabled, or constrained with an explicit thinking budget. The same checkpoint can therefore serve direct answers and planned coding work.
NVIDIA lists German, English, French, and Japanese among the supported languages. The model card specifies 9 billion parameters and up to 128K context tokens. Its family successor, Nemotron 3 Nano, carries the hybrid design into a larger sparse MoE architecture.
Specifications and access
| Specification | Value and source |
|---|---|
| Weights | Open weightsSource |
| Access | API, LocalSource |
| Parameters | 9BSource |
| Context window | 128,000 tokensSource |
| Knowledge cutoff | September 2024Source |
| Notes | Open-weights reasoning and coding model with 9B parameters and a 128K context window.Source |
| Release | 2025-08-18Source |
| Parameters | 9 BSource |
| License | NVIDIA Open Model LicenseSource |
| Architecture | Mamba2-Transformer hybrid (Nemotron-Hybrid)Source |
| Languages | English, German, Spanish, French, Italian, JapaneseSource |
| Input | TextSource |
| Output | TextSource |
Page 1 of 2
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.
AIME25 (Show measurement, test conditions, and source)
- Source value
- 72.1
- Score
- 72.1
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Reasoning-On
BFCL v3 (Show measurement, test conditions, and source)
- Source value
- 66.9
- Score
- 66.9
- Metric
- percent
- Unit
- %
- Category
- tool-use
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Reasoning-On
GPQA (Show measurement, test conditions, and source)
- Source value
- 64
- Score
- 64
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Reasoning-On
LCB (Show measurement, test conditions, and source)
- Source value
- 71.1
- Score
- 71.1
- Metric
- percent
- Unit
- %
- Category
- coding
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Reasoning-On
MATH500 (Show measurement, test conditions, and source)
- Source value
- 97.8
- Score
- 97.8
- Metric
- percent
- Unit
- %
- Category
- reasoning
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Reasoning-On
RULER (128K) (Show measurement, test conditions, and source)
- Source value
- 78.9
- Score
- 78.9
- Metric
- percent
- Unit
- %
- Category
- long-context
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- NVIDIA model card
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Reasoning-Off
Artificial Analysis Intelligence Index (Show measurement, test conditions, and source)
- Source value
- 6.84412514839364
- Score
- 6.84412514839364
- Metric
- points
- Unit
- points
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. Non-reasoning
AA-Omniscience Index (Show measurement, test conditions, and source)
- Source value
- -57.25
- Score
- -57.25
- Metric
- points
- Unit
- points
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. Non-reasoning
Output speed (Show measurement, test conditions, and source)
- Source value
- 143.083423156384
- Score
- 143.083423156384
- Metric
- tokens/s
- Unit
- tokens/s
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. Non-reasoning
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median.
Latency reasoning time (Show measurement, test conditions, and source)
- Source value
- 0
- Score
- 0
- Metric
- seconds
- Unit
- seconds
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. Non-reasoning
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median.
Latency input time (Show measurement, test conditions, and source)
- Source value
- 2.02425794000002
- Score
- 2.02425794000002
- Metric
- seconds
- Unit
- seconds
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. Non-reasoning
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median.
Answer generation time (end-to-end timing breakdown) (Show measurement, test conditions, and source)
- Source value
- 3.494464900057092
- Score
- 3.494464900057092
- Metric
- seconds
- Unit
- seconds
- Direction
- Lower is better
- Source type
- independent-evaluator
- Evaluator
- Artificial Analysis
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Independent evaluation. Non-reasoning
- Context
- Artificial Analysis performance timing; output speed is decoded after the first streaming chunk. Provider basis: serving-provider median.
Page 1 of 2
Popularity
44.8 / 100
Predecessor and successor
Sources and data date
| Type | Evidence and data date |
|---|---|
| Speed | Median across serving providers, Non-reasoning. Output speed after the first streaming chunk, not total response time. Retrieved September 8, 2026. |
| Price | Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. |
| Input | Documented modalities. Retrieved September 8, 2026. |
| Output | Documented modalities. Retrieved September 8, 2026. |
| Research status | Research date September 8, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | active-parameters-billions Original wording No exact value located in the checked primary model card/API documentation. |
| Not found in the inspected sources | knowledge-cutoff Original wording No exact value located in the checked primary model card/API documentation. |
| Additional source | Model metadata |
| Additional source | NVIDIA Nemotron 3 Nano model card · Editorial description reviewed October 2, 2026 |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Creative Commons Attribution 4.0 |
| Additional source | NVIDIA Nemotron 3 Nano launch blog (Hugging Face) (2026-09-19) |