NVIDIA Nemotron 3 Ultra vs Step 5 Preview
NVIDIA · US | StepFun · China · Updated June 2026
Quick verdict
Pick NVIDIA Nemotron 3 Ultra for the most capable open-weight model from a us lab (artificial analysis intelligence index of about 48) or fast, efficient long-horizon agentic reasoning via a hybrid mamba-transformer design. Pick Step 5 Preview for sparse moe architecture — 600b total parameters, only 27b active per token — built for long-horizon agentic work across large codebases and documents or 1m-token context window (64k max output) at $1/$2.70 per million tokens, among the cheapest frontier-scale context windows available. Choose NVIDIA Nemotron 3 Ultra if you need self-hosting or data privacy; Step 5 Preview if you want a managed API.
NVIDIA Nemotron 3 Ultra (NVIDIA, US) and Step 5 Preview (StepFun, China) line up two different AI ecosystems against each other — a comparison that is as much about cost philosophy and openness as raw capability. NVIDIA Nemotron 3 Ultra is nVIDIA's open-weight reasoning flagship (about 550B total, 55B active) — the most capable open model from a US lab, built for long-running agents. Step 5 Preview is stepFun's September 20, 2026 flagship preview — a 600B-parameter sparse MoE model (27B active) with a 1M-token context window at $1/$2.70 per million tokens, open weights promised for mid-October. They diverge most on price and open vs. closed weights — each quantified below from the models' real specs.
Key differences at a glance
- ▸Cost model: NVIDIA Nemotron 3 Ultra ships open weights you can self-host (hardware cost only, no per-token fee), while Step 5 Preview is API-metered at $1/$2.7 per 1M tokens. Your choice depends on whether you want zero marginal cost at the price of running infrastructure.
- ▸Context window: both advertise 1M (~1,500 pages). Tie on paper — test on your own long inputs, since usable recall varies by model.
- ▸Recency: Step 5 Preview is the newer model by about 4 months (released September 20, 2026), usually meaning fresher training data and capabilities.
- ▸Ecosystem: this is a US-vs-China matchup — they differ in pricing philosophy, data-residency options, and tooling ecosystems, not only benchmarks.
Side-by-side specs
| Spec | NVIDIA Nemotron 3 Ultra | Step 5 Preview |
|---|---|---|
| Provider | NVIDIA (US) | StepFun (China) |
| Released | June 4, 2026 | September 20, 2026 |
| Context window | 1M (~1,500 pages) | 1M tokens (~1,500 pages) |
| Price (in/out) | Open weight (self-host / free) | $1/$2.7 per 1M tokens |
| Open weight? | Yes — self-hostable | No — API only |
| Modalities | text, code | text, image, code |
| SWE-Bench Verified | Not published | Not published |
| MRCR v2 @ 1M | Not published | Not published |
Who wins what
The most capable open-weight model from a US lab (Artificial Analysis Intelligence Index of about 48)
NVIDIA Nemotron 3 Ultra
Open weights make this possible at all — Step 5 Preview is API-only, so it cannot leave the vendor's servers.
Fast, efficient long-horizon agentic reasoning via a hybrid Mamba-Transformer design
NVIDIA Nemotron 3 Ultra
NVIDIA's open-weight reasoning flagship (about 550B total, 55B active) — the most capable open model from a US lab, built for long-running agents — and its weights are open while Step 5 Preview is API-only.
A fully open release — weights, training data, and recipes under a permissive license
NVIDIA Nemotron 3 Ultra
Step 5 Preview is comparatively weak here — open weights were promised for October 15, 2026 but hadn't shipped as of this writing — API-only for now despite the open-weight roadmap
Sparse MoE architecture — 600B total parameters, only 27B active per token — built for long-horizon agentic work across large codebases and documents
Step 5 Preview
StepFun's September 20, 2026 flagship preview — a 600B-parameter sparse MoE model (27B active) with a 1M-token context window at $1/$2.70 per million tokens, open weights promised for mid-October — and it is the newer of the two.
1M-token context window (64K max output) at $1/$2.70 per million tokens, among the cheapest frontier-scale context windows available
Step 5 Preview
NVIDIA Nemotron 3 Ultra is comparatively weak here — a 550B mixture-of-experts is heavy to self-host, and the 1M context is rarely served in full
67.7% on DeepSWE v1.1, ahead of open-weight rivals Kimi K3 (67.5%) and GLM-5.3 (66.9%)
Step 5 Preview
Step 5 Preview lists 67.7% on DeepSWE v1.1, ahead of open-weight rivals Kimi K3 (67.5%) and GLM-5.3 (66.9%) among its strengths; NVIDIA Nemotron 3 Ultra does not.
Lowest cost at scale
NVIDIA Nemotron 3 Ultra
Its weights are open, so at volume you pay for your own hardware instead of Step 5 Preview's $1/$2.7 per 1M tokens.
Which should you pick?
A cost-sensitive startup shipping high volume
→ NVIDIA Nemotron 3 Ultra
At Open weight (self-host / free) it undercuts Step 5 Preview, and on millions of tokens that margin decides the monthly bill.
A team with data-privacy or self-hosting needs
→ NVIDIA Nemotron 3 Ultra
Open weights let you run it on your own hardware; Step 5 Preview is API-only.
Anyone whose priority is the most capable open-weight model from a us lab (artificial analysis intelligence index of about 48)
→ NVIDIA Nemotron 3 Ultra
It is specifically built for that.
Anyone whose priority is sparse moe architecture — 600b total parameters, only 27b active per token — built for long-horizon agentic work across large codebases and documents
→ Step 5 Preview
That is its strongest area.
An enterprise with regional data-residency rules
→ NVIDIA Nemotron 3 Ultra or Step 5 Preview
Origin (US vs China) affects where data is processed and which compliance regime applies — check the provider's terms for your region.
NVIDIA Nemotron 3 Ultra: where it fits
NVIDIA's open-weight reasoning flagship (about 550B total, 55B active) — the most capable open model from a US lab, built for long-running agents. Released June 4, 2026 by NVIDIA, it is built for the most capable open-weight model from a US lab (Artificial Analysis Intelligence Index of about 48), fast, efficient long-horizon agentic reasoning via a hybrid Mamba-Transformer design, a fully open release — weights, training data, and recipes under a permissive license, and strong coding for an open model (SWE-Bench Verified in the high 60s).
Its trade-offs are real: trails the best Chinese open models on overall intelligence, and a 550B mixture-of-experts is heavy to self-host, and the 1M context is rarely served in full. As an open-weight model, its running cost is your own hardware rather than a per-token fee.
Step 5 Preview: where it fits
StepFun's September 20, 2026 flagship preview — a 600B-parameter sparse MoE model (27B active) with a 1M-token context window at $1/$2.70 per million tokens, open weights promised for mid-October. Released September 20, 2026 by StepFun, it is built for sparse MoE architecture — 600B total parameters, only 27B active per token — built for long-horizon agentic work across large codebases and documents, 1M-token context window (64K max output) at $1/$2.70 per million tokens, among the cheapest frontier-scale context windows available, 67.7% on DeepSWE v1.1, ahead of open-weight rivals Kimi K3 (67.5%) and GLM-5.3 (66.9%), and scores 44 on the Artificial Analysis Intelligence Index, ranking 27th of 653 models tracked.
Its trade-offs: trails closed frontier models on coding — 67.7% on DeepSWE v1.1 vs GPT-6 Astra's 74.1% and Claude Opus 5's 74.0%, open weights were promised for October 15, 2026 but hadn't shipped as of this writing — API-only for now despite the open-weight roadmap, a 'Preview' release, not StepFun's finished flagship — positioning and pricing may change once weights ship, and stepCodeBench and DeepSWE figures are StepFun's own reported numbers, not yet independently reproduced at scale. At $1 in / $2.7 out per million tokens, it sits in the budget price band.
The bottom line for this matchup
The defining split here is open vs. closed. NVIDIA Nemotron 3 Ultra gives you weights you control — self-host it, fine-tune it, keep data in-house, pay only for hardware. Step 5 Preview gives you a managed, always-updated API with no infrastructure to run. Teams with GPUs, privacy requirements, or huge volume often favour the open model; teams that want zero ops and the latest capabilities favour the closed one. Capability is close enough that this operational question, not the benchmark, usually decides it.
Want both NVIDIA Nemotron 3 Ultra and Step 5 Preview without two subscriptions? LumiChats gives you these plus 40+ models under one ₹69/day pass (about $1/day) — draft with one, cross-check with the other.
See pricingFrequently asked questions
Is NVIDIA Nemotron 3 Ultra or Step 5 Preview better for coding?
Public SWE-Bench figures are not available for either model, so the honest test is your own repository — run an identical real bug through both. By design, NVIDIA Nemotron 3 Ultra leans toward the most capable open-weight model from a us lab (artificial analysis intelligence index of about 48) while Step 5 Preview leans toward sparse moe architecture — 600b total parameters, only 27b active per token — built for long-horizon agentic work across large codebases and documents, and that positioning usually predicts which feels better on your codebase.
Which is cheaper, NVIDIA Nemotron 3 Ultra or Step 5 Preview?
NVIDIA Nemotron 3 Ultra is open-weight, so self-hosting means no per-token fee (you pay for hardware instead), while Step 5 Preview is API-metered at $1/$2.7 per 1M tokens. For most teams without GPUs, the API model is cheaper to start; at very high volume, self-hosting can win.
Which has the bigger context window?
Both advertise 1M (~1,500 pages). Remember advertised ≠ usable: recall typically degrades before the ceiling.
Can I use both NVIDIA Nemotron 3 Ultra and Step 5 Preview together?
Yes — a multi-model platform like LumiChats gives you NVIDIA Nemotron 3 Ultra, Step 5 Preview and 40+ others under one ₹69/day pass (about $1/day), so you can draft with one and cross-check with the other instead of buying two subscriptions.
Which is newer, NVIDIA Nemotron 3 Ultra or Step 5 Preview?
Step 5 Preview — released September 20, 2026, about 4 months after NVIDIA Nemotron 3 Ultra.
Related comparisons
Specifications and benchmarks reflect publicly reported figures as of June 2026 and may change as providers release updates. Always verify on your own workload.