> Nebius
Nebius sells two products relevant to an API-pricing decision: Token Factory, an OpenAI-compatible inference API for open models (where each model is offered in a default and a -fast flavor with different latency and token pricing), and GPU instances billed per GPU-hour, from $2.15 for preemptible H100 up to $7.85 on-demand for HGX B300. The public pricing page documents instance, storage and network rates plus billing rules; Token Factory per-token prices are only shown after login. Commitment reservations are advertised at up to 35% below on-demand rates.
// pros
- Token Factory limit scaling is documented in numbers: from a baseline example of 60 RPM / 400,000 TPM, limits grow 20% for each rolling 15-minute window at 80% or more utilisation and can reach 20x the base allocation before Enterprise is required.
- Each model comes in two flavors (base and -fast, selected by appending -fast to the model name) with identical model outputs but different token pricing and latency.
- Preemptible pricing is roughly half of on-demand across the GPU table: HGX H100 $2.15 vs $3.85 per GPU-hour, HGX H200 $2.45 vs $4.50, RTX PRO 6000 $0.95 vs $1.80.
- Managed Service for Kubernetes, Soperator (Slurm on Kubernetes software) and networking ingress/egress plus public IP addresses are listed as free, with object-storage egress free on the Intelligent and Enhanced classes.
- Response headers expose the live dynamic scale factor (x-ratelimit-dynamic-scale-requests, -tokens) and current window usage, so quota growth is observable rather than guessed.
- Batch API is documented as having significantly higher limits than the synchronous path, useful for large offline jobs.
// cons
- No standing free tier: setting up billing charges $25 and credits it to the account balance, and free credits otherwise arrive only through a promo code that Nebius issues in a special offer.
- Token Factory per-token prices sit behind a login (the prices page requires an account), so official public pages cannot be used to compare model rates without signing up.
- The largest clusters (GB300 NVL72, GB200 NVL72) carry no public price and are contact-sales only, and commitment discounts require a company account and a sales contact.
- Individuals can pay only by card and can update card details only in the web console; bank transfer is a company-only option routed via sales.
- Base rate limits are soft caps: bursts may be throttled and exceeding the current limit returns HTTP 429 even while the limit auto-scales.
Choose Nebius when you need GPU capacity and an OpenAI-compatible inference endpoint from one account and value automatic limit scaling over a published free tier. Treat it as the paid backend behind a free-tier-first router, and check token cost inside the console because public per-token rates are not exposed.
written by Mike Fleming, Uprouter editorial · verified against official sources September 16, 2026
Free tier present — value not quantified
This provider offers a free tier, but its terms don’t convert cleanly into a USD figure (e.g. request-count limits or capacity-dependent pools). We refuse to print a misleading zero here. See the note below and the official pricing page for the exact limits.
note: No free tier or USD sign-up credit is stated on official pages (checked 2026-09-16); the docs describe promo codes that grant free credits only when Nebius issues one in a special offer, and the first card payment is $25 credited to the account balance.
| Plan | Type | Monthly | Input /1M | Output /1M | Note |
|---|---|---|---|---|---|
| Free tier (source-reported) | free | — | — | — | New users get free credit (reports vary from $1 to $100, no credit card needed); additionally AI Studio offers an ongoing free API for models like Meta-Llama-3.3-70B-Instruct and Qwen3-235B-A22B with tier-based rate limits. |
| Pay-as-you-go | payg | — | $0.200 | $0.600 | Pricing for Llama 3.1 70B based on community reports and snippet data. |
Pricing normalized from public sources — always verify with the provider.
// model prices
| Model | Input /1M | Output /1M | Context |
|---|---|---|---|
| DeepSeek family | — | — | — |
| Llama/Qwen and 60+ open-source models | — | — | — |
| Qwen3-235B-A22B | — | — | 131k |
| Llama 3.3 70B Instruct | — | — | 131k |
$ Does Nebius have a free tier?
Yes — Nebius offers a free tier. Always confirm current limits on Nebius's official pricing page — free tiers change without notice.
$ How much does Nebius cost per 1M tokens?
Nebius does not publish simple per-1M-token pricing (it may bill per request, per output, or via a subscription plan). Check their official pricing page for exact figures.
$ Is Nebius safe to use?
Uprouter rates Nebius as low risk. Medium. There is no free tier to fall back on: the $25 first card payment becomes account balance, so usage is prepaid spend from the start, and the cheapest public entry point is $0.90/GPU-hour preemptible L40S. Token Factory pricing is visible only after login, which makes budgets harder to verify before committing. Nebius also warns of an operational change in this area: its own notice states that from 21 September 2026 the primary Token Factory storage location moves from Finland to France with a temporary read-only period. Risk ratings are editorial, evidence-based, and never influenced by affiliate relationships.
Raw community sentiment, shown unblended. Votes never feed into objective data fields like price, risk, or status — and never into sort order.
No published reviews yet — be the first.
// more_informationprice history · live status · risk & confidence · code snippets
price_history
No recorded changes yet.
live_status
full history →best-effort probes · not a guarantee · last 15h ago
| Checked | Status | Latency |
|---|---|---|
| 15h ago | up | 345ms |
| 1d ago | up | 21ms |
| 3d ago | up | 785ms |
| 4d ago | up | 441ms |
| 6d ago | up | 651ms |
| 7d ago | up | 1039ms |
| 9d ago | up | 69ms |
| 10d ago | up | 29ms |
| 12d ago | up | 85ms |
| 13d ago | up | 1030ms |
| 14d ago | up | 52ms |
| 15d ago | up | 2994ms |
| 15d ago | up | 1013ms |
| 16d ago | up | 1291ms |
| 17d ago | up | 75ms |
| 18d ago | up | 93ms |
| 18d ago | up | 77ms |
| 19d ago | up | 22ms |
| 20d ago | up | 1010ms |
| 21d ago | up | 1023ms |
| 22d ago | up | 898ms |
| 22d ago | up | 910ms |
| 23d ago | up | 24ms |
| 24d ago | up | 1010ms |
| 25d ago | up | 63ms |
| 26d ago | up | 1061ms |
| 26d ago | up | 1068ms |
| 27d ago | up | 26ms |
| 28d ago | up | 1222ms |
| 29d ago | up | 1116ms |
Uprouter runs its own probes; best-effort, not a guarantee. Probe results depend on our network location and cadence — for production, always use the provider’s own status page.
risk_&_confidence
Medium. There is no free tier to fall back on: the $25 first card payment becomes account balance, so usage is prepaid spend from the start, and the cheapest public entry point is $0.90/GPU-hour preemptible L40S. Token Factory pricing is visible only after login, which makes budgets harder to verify before committing. Nebius also warns of an operational change in this area: its own notice states that from 21 September 2026 the primary Token Factory storage location moves from Finland to France with a temporary read-only period.
Confidence reflects how much verifiable evidence (official docs, probe history, community reports) backs this entry. Risk is our editorial assessment of operator reliability and terms — not financial advice.
code_snippets
# Nebius's own API — get a key from their dashboard
curl https://YOUR-PROVIDER-ENDPOINT/chat/completions \
-H "Authorization: Bearer THEIR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-family",
"messages": [{ "role": "user", "content": "Hello" }]
}'Direct to the provider — Uprouter receives nothing; verify their current pricing before purchasing.
Ranked by neutral similarity signals only — live status, free-tier shape and free-credit proximity. Commercial relationships never influence this list.