NVIDIA Nemotron 3.5 Lightning
by DeepInfra · 30B (3B active) parameters
Today's price · per 1M tokens
Input
$0.060
per 1M tokens
Output
$0.160
per 1M tokens
Blended
$0.085
blended $/1M — 3:1 weighted input:output
Cached input
$0.030
50% of input — prompt caching
How this price is scoped: DeepInfra's own list price, per 1M tokens: deepinfra.com/pricing prints NVIDIA-Nemotron-3.5-Lightning at 256k context, "$0.06 / $0.03 cached" input and "$0.16" output, and DeepInfra's model API returns the same rates. Until October 2026 DeepInfra listed it at $0.08 input, $0.04 cached and $0.20 output. Read 2026-10-18 and 2026-10-19.
Price receipt
We read DeepInfra's own pricing page on and found NVIDIA-Nemotron-3.5-Lightning listed with a price on the same row — that read is where the number above comes from. We keep the page text we read; its content hash is 30a228bbd8.
Overview
NVIDIA Nemotron 3.5 Lightning served on DeepInfra at $0.06/$0.16 per 1M tokens, with cached input at $0.03. The cheapest of the three hosts serving it (CoreWeave and Venice both list $0.10/$0.25). NVIDIA publishes no per-token price of its own for this model, so a host rate is the only price it has.
Deploy this open model on rented GPUs
Open weights mean you can self-host instead of paying per-token API prices. These platforms let you serve it on demand — often cheaper at scale.
Partner links — we may earn a commission if you sign up, at no cost to you. We only list platforms we'd recommend regardless.
Capabilities
struck through = not supportedBenchmark performance
accuracy % · higher is betterSources: NVIDIA NIM API docs model card (NVFP4 variant) (GPQA Diamond, BrowseComp) · NVIDIA NIM API docs model card (NVFP4 variant, SWE-bench Verified) (SWE-Bench Verified) · NVIDIA NIM API docs model card (Terminal-Bench 2.1, NVFP4 variant) (Terminal-Bench 2.1) · NVIDIA NIM API docs model card (HLE text-only, no tools, NVFP4 variant) (Humanity's Last Exam)
Every per-benchmark score we hold for this model is reported by its own vendor, so none is counted toward an accuracy average or any ranking. The percentile above is its standing across the independent composites below.
- AA Intelligence Index: 5.5 (Artificial Analysis composite across reasoning, knowledge and coding evals)
- AA-Omniscience Index: -15.7 (−100–100)
- GDPval-AA v2: 645.3 ELO (Real-world work tasks, human baseline = 1000)
- Humanity's Last Exam: 10.6%
- GPQA Diamond: 74.3%
- AA-LCR (long-context reasoning): 60.3%
- CritPt (research-level physics): 0%
- Terminal-Bench 2.1: 24.3%
- Terminal-Bench 4.0: 0.5%
- SciCode: 32.1%
- Cost to run the AA Intelligence Index: $0.314 per task, 4.3 min per task
- Measured by AA: 254.3 tokens/sec output, 0.88s to first token, 14.19s to first answer token
- Ranks #162 of 189 comparably-measured models by percentile score, across 3 independent measurements
- Ranks #13 of 189 comparably-tested models by normalized performance per dollar
254.3 tokens/sec output 0.88s latency to first token (TTFT)
Throughput: Artificial Analysis, measured 2026-11-14 · Source: Artificial Analysis (artificialanalysis.ai)
Source: Artificial Analysis, NVIDIA NIM API docs model card (HLE text-only, no tools, NVFP4 variant), NVIDIA NIM API docs model card (NVFP4 variant) · updated · See full rankings →
Specifications
- Provider
- DeepInfra
- Context window
- 256K tokens
- Modality
- text
- Parameters
- 30B (3B active)
- Open source
- Yes — open weights available
- Released
- Status
- Current
- Last updated
- Tags
Availability verified: — listed on DeepInfra's own page