NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 is an open-weight language model from NVIDIA with 31.6B parameters (mixture of experts). At 4-bit it needs about 21.3 GB of memory, so the smallest current device that fits it is the Arc Pro B60.
Overview
Includes room for an 8k context and runtime overhead. More context needs more memory.
Key information
- Lab
- NVIDIA
- Released
- 1 Aug 2026
- Parameters
- 31.6B · MoE
- Type
- Language
- Context
- 262,144 tokens
- License
- Custom license
- HF downloads, 30 days
- 995,986
More from NVIDIA
View allWill it run on your hardware?
How we estimate| Estimate or device | Memory |
|---|---|
| Q4 | 21.3 GB |
| Q8 | 37.3 GB |
| BF16 | 70 GB |
| RTX 5060 Ti 16 GB | 16 GB usable |
| RTX 5090 32 GB | 32 GB usable |
| DGX Spark 128 GB | 108.8 GB usable |
| Mac Studio M5 Ultra 256 GB | 192 GB usable |
Single-device memory estimate. Pick a device to update the On your device column.
36 GB usable memory
| Format | Memory needed | On your device | Example hardware |
|---|---|---|---|
| 4-bit (Q4) | ≈ 21.3 GB | Fits 59% | Arc Pro B6024 GB memory |
| 8-bit (Q8) | ≈ 37.3 GB | Too big 104% | L40S48 GB memory |
| 16-bit (BF16) | ≈ 70 GB | Too big 194% | A100 80GB SXM80 GB memory |
Fits means the estimate stays within 90% of usable memory. Tight is above 90% and up to 100%. Too big is above 100%. These are memory checks, not speed or compatibility tests.
Mac allowances are 66% of total RAM up to 36 GB, and 75% above. AI boxes allow 85% because their memory is shared with the operating system.
Where to run it
On a Mac
Start with the MLX files, built for Apple Silicon.
mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-bf16Check your MacThrough an API
Darkbloom (int4) is the cheapest provider we found.
$0.025 / $0.12
Input / output per 1M tokens
₹2.36 / ₹11 at ₹94.29 per $1
Blended: $0.0488 (₹4.60) per 1M. Ranked by a 3:1 mix of input and output tokens.
mlx_lm.generate --model mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-bf16 \
--prompt "Explain how a transformer works."Example commands built from the repository ids above. They need a working install of MLX. We have not run them; check each repository for available files.
| Precision | |||||
|---|---|---|---|---|---|
| DarkbloomCheapest | $0.0488₹4.60 | $0.025₹2.36 | $0.12₹11 | 262,144 | INT4 |
| DeepInfra | $0.085₹8.01 | $0.06₹5.66 | $0.16₹15 | 262,144 | BF16 |
| Fireworks | $0.0875₹8.25 | $0.05₹4.71 | $0.20₹19 | 262,144 | Not listed |
| CoreWeave | $0.1025₹9.66 | $0.07₹6.60 | $0.20₹19 | 262,144 | BF16 |
INR at ₹94.29 per $1. Prices fetched 14 Nov 2026.
More about this model
Frequently asked questions
How much memory does NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 need? At an 8k context, about 21.3 GB at 4-bit, 37.3 GB at 8-bit and 70 GB at 16-bit (BF16). A longer context needs more.
Can I run NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 on a Mac? The smallest current Mac that fits the 4-bit estimate of 21.3 GB is the Mac Studio M5 Max (36GB). An MLX build is available: mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-bf16.
Does the full context fit? NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 supports up to 262,144 tokens of context. Our memory estimates use 8k, so long prompts need more memory than shown.
What does it cost through an API? The cheapest provider we found is Darkbloom, at $0.03 per 1M input tokens and $0.12 per 1M output tokens (₹2.36 and ₹11 at 94.29 rupees per dollar). Ranked by a 3:1 mix of input and output tokens.
License in plain English
Custom license. This model uses a custom license rather than a standard one. Read it on the model card before using the model commercially. This summary is not legal advice.
OMHQ RigBench
Not yet tested. We show no speed or quality score without a measurement.
Revision history and sources
Last updated . Sources:Hugging Face model card, MLX repository, OpenRouter.