Nebius

nebius.com · Compute · 4 records · JSON

Nebius AI Cloud Compute★ qualifiescard?cardphoneM
Nebius ConTree (Token Factory Sandboxes) Computecaveatemail?card?Dwrite-gated
Nebius Computepaidemail?card?Drequest-rate
Nebius research grants Computefails: identityinstitutionno cardDwrite-gated

Nebius AI Cloud Compute

★ qualifiescard?cardphoneM

nebius.com/spot-prices · GPU cloud · compute/nebius-ai-cloud · JSON · compare

Free tier	The deepest H100 discount found anywhere in this class, read live and anonymously. NVIDIA H100 NVLink in eu-north1 at USD 0.79 per GPU-hour against USD 4.50 on-demand, which is 17.6 per cent of on-demand, a 5.70 times reduction. The same page at the same moment shows H200 at USD 0.89 in us-central1 against USD 5.40 on-demand (16.5 per cent, 6.07 times), B200 at USD 3.60 in us-central1, and B300 at USD 5.89 in eu-west2 and us-north1. The H100 price had been effective since 9 October 2026 17:01, so it was stable for 13 days at the time of reading. unit hours; metered on wall-clock; none
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	card (placeholder) unverified: no page stating the signup requirement was read; the record's evidence covers the spot ladder (H100 at USD 0.79 per GPU-hour against USD 4.50 on-demand), the 15-minute repricing rule, the one-cent-below-on-demand ceiling, 60-second SIGTERM preemption, and that prices are viewable with no account. Whether signup needs a card, a phone number or an ID document, and whether any hold is taken, is not stated.
Binding	none The price is the floor of a dynamic range that can rise to one cent below on-demand every 15 minutes, and the deep rates are single-region, so the failure mode is the rate moving rather than the capacity being unreachable.
Interface	rest inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	USD 0.79 an H100-hour means USD 300 of credit buys 380 H100-hours, which is the largest measured H100 budget in the corpus at any identity cost.
Reliability	no reliability event recorded
Notes	This number is better than every third-party H100 spot rate found and it is measurable with no account, which makes it the reference point the rest of the class should be judged against. It is also 2.8 times cheaper than Verda's flat 50 per cent H100 spot at USD 2.235. The risk is entirely in the dynamic clause: the range ceiling is one cent under USD 4.50, so a run without a pricing policy has unbounded cost, and a pricing policy that caps it converts the price risk into a stop. H100 is single-region (eu-north1), so there is no region to move to when the price rises. qualifies-with-caveat via a05-spot. The one recorded is the better-evidenced.
Evidence rung	measured
Checked	2026-10-22 (live); stale on 2026-12-06

Evidence (11)

date	source	what it proved
2026-10-22	nebius.com/spot-prices	probe.py record 20261022T134509, status 200, 202986 bytes, sha256 1dff7ee3cbb2c33999164e31daf37bd124d2d68f1a0b0774cf1e630b86ae5d8f, page states 'Updated 22 Oct 2026, 13:40 UTC'
2026-10-22	nebius.com/spot-prices	the All live prices table lists 'NVIDIA H100 \| eu-north1 \| $0.79 \| 9 Oct 2026, 17:01', 'NVIDIA H200 \| us-central1 \| $0.89 \| 22 Oct 2026, 13:11', 'NVIDIA B200 \| us-central1 \| $3.60', and 12 markets in total
2026-10-22	docs.nebius.com/compute/resources/pricing	on-demand NVIDIA H100 NVLink is USD 4.50 per GPU hour and H200 NVLink is USD 5.40, both from October 8 2026, quoted from the pricing tables
2026-10-22	docs.nebius.com/compute/resources/pricing	the change note states that from October 8 2026 preemptible pricing for B300, B200, H200, H100 and RTX PRO 6000 switches to dynamic spot pricing and the preemptible cell shows the minimum spot price, quoted
2026-10-22	docs.nebius.com/compute/resources/pricing	the B300 table gives USD 9.50 on-demand and 'From $0.99' preemptible from October 8 2026, USD 4.30 preemptible between October 1 and October 8, and the H200 table gives USD 5.40 on-demand against 'From $0.79'
2026-10-22	docs.nebius.com/signup-billing/pricing-policy	the spot price 'may change as often as every 15 minutes', 'The highest preemptible VM price is always one cent below the current pay-as-you-go (PAYG) on-demand rate', and spot prices are viewable without an account
2026-10-22	nebius.com/spot-prices	the All live prices table gives three H200 regions at USD 0.89, USD 2.45 and USD 5.09 on the same reading, and RTX PRO 6000 at USD 1.08, USD 1.38 and USD 1.39, read at 13:40 UTC
2026-10-22	nebius.com/spot-prices	the per-platform cards show 24-hour movement: H200 us-central1 down USD 0.41 or 31.5 per cent, B300 eu-west2 up USD 1.80 or 44.0 per cent, RTX PRO 6000 uk-south2 down USD 0.30 or 21.7 per cent
2026-10-22	docs.nebius.com/signup-billing/pricing-policy	'You don't need a Nebius AI Cloud account to view them', quoted, and the worked example charges 8 GPUs at USD 2.00 for the first 15 minutes and USD 2.50 for the next
2026-10-22	docs.nebius.com/signup-billing/pricing-policy	the 15-minute interval billing example: 8 GPUs at 0.25 hours at USD 2.00 is USD 4.00, then 8 GPUs at 0.25 hours at USD 2.50 is USD 5.00
2026-10-22	docs.nebius.com/compute/virtual-machines/preemptible	'Compute sends a SIGTERM signal 60 seconds before stopping the VM', volumes are preserved, and the only supported preemption action is STOP, quoted

More fields (1)

{
 "wave2": {
  "class": "spot-arbitrage",
  "free_tier_kind": "none",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": "H100-80GiB",
   "gpu_count": 1
  }
 }
}

Unknown or withheld values (9)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.kind identity-defaulted
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated

Nebius ConTree (Token Factory Sandboxes) Compute

caveatemail?card?Dwrite-gated

tokenfactory.nebius.com/sandboxes/about · container and VM sandbox · compute/nebius-contree · JSON · compare

Free tier	free while in beta: runs do not consume credits, but calling it from your own agents requires beta access unit none; metered on none
Price	none stated
Per-object cap	not stated
Retention	not stated
Egress	not stated
Identity	email (placeholder) not published
Limits	write-gated calling the sandbox from your own agents requires beta access; the browser shell is open source retention-clamped the free position is 'while in beta', so it ends when beta does and no notice is promised source not stated
Binding	write-gated
Interface	unknown
Tooling	not stated
Composes with	not stated
Reliability	no reliability event recorded
Evidence rung	documented the sandboxes page was read on 2026-10-22 and states the free-while-in-beta condition and the access gate.
Checked	2026-10-23 (live); stale on 2026-11-22

Evidence (2)

date	source	what it proved
2026-10-02	tokenfactory.nebius.com/sandboxes/about	not published
2026-10-22	tokenfactory.nebius.com/sandboxes/about	'Free while in beta - runs don't consume your credits.' and 'request beta access to call it from your own agents - free while in beta', quoted

More fields (14)

{
 "talaria_m10h_usd": null,
 "talaria_m24h_usd": null,
 "nemo_m10h_usd": null,
 "keep_rate": null,
 "keep_rate_basis": "no field in the source",
 "billing_model": "unclassified",
 "shape": null,
 "session_cap": {
  "wall_clock_min": null,
  "idle_reap_s": null,
  "restart_required": false,
  "note": "the page states no session clock; the gate is beta access rather than a quota"
 },
 "persistent_storage": null,
 "network_out": null,
 "price_note": null,
 "free_tier_kind": "allowance",
 "ps": null,
 "net": {
  "out": true,
  "in": false,
  "ports": "SDK and an in-browser shell",
  "static_ip": false,
  "ipv6": null,
  "allowance": null
 }
}

Unknown or withheld values (16)

$.chaining not-stated
$.egress.metered egress-unknown
$.egress.terms not-stated
$.evidence[0].proved process
$.extras.price_note process
$.identity_cost.detail defaulted
$.identity_cost.kind identity-defaulted
$.interface interface-unknown
$.interface_detail withheld
$.limits_note not-stated
$.per_object_cap.notes process
$.per_object_cap.stated not-stated
$.reliability_note none-observed
$.retention.notes not-extracted
$.retention.stated not-stated
$.tooling not-stated

Nebius Compute

paidemail?card?Drequest-rate

docs.nebius.com/compute/resources/pricing · GPU cloud · compute/nebius · JSON · compare

Free tier	no free tier: computing resources bill per second of running time only, with a B300 NVLink GPU at $9.50 a GPU-hour on demand, and a stopped VM not charged for compute unit none; metered on none
Price	none stated
Per-object cap	not stated
Retention	not stated
Egress	not stated
Identity	email (placeholder) not published
Limits	request-rate $9.50 a GPU-hour for a B300 NVLink on demand; billed per second with a one-hour pricing unit source retention-clamped a stopped VM is not charged for compute but its storage volumes are; preemptible pricing for the newest GPU platforms switched to dynamic spot pricing on 2026-10-08 source not stated
Binding	request-rate
Interface	unknown
Tooling	not stated
Composes with	not stated
Reliability	no reliability event recorded
Evidence rung	documented the GPU pricing page was read on 2026-10-22 and states the billing and pricing units, the stopped-VM rule and the B300 rates. There is no free tier, so the verdict moves to paid-only
Checked	2026-10-23 (live); stale on 2026-11-22

Evidence (2)

date	source	what it proved
2026-10-02	docs.nebius.com/compute/resources/pricing	not published
2026-10-22	docs.nebius.com/compute/resources/pricing	'You are charged for computing resources (GPUs, vCPUs, RAM) of running virtual machines (VMs). Computing resources of stopped VMs are not charged (this does not apply to storage volumes)' with 'Billing unit: 1 second' and 'Pricing unit: 1 hour', and B300 NVLink at $9.50 a GPU-hour or $7.85 on a different term, quoted

More fields (15)

{
 "talaria_m10h_usd": 11.04,
 "talaria_m24h_usd": 26.5,
 "nemo_m10h_usd": 15.46,
 "keep_rate": 1.0,
 "keep_rate_basis": null,
 "billing_model": "unclassified",
 "shape": {
  "vcpu": null,
  "ram": null,
  "disk": null,
  "gpu_class": "NVIDIA B300 NVLink",
  "gpu_count": 8,
  "note": "the gpu-b300-sxm platform; preemptible B300 starts from $0.99 a GPU-hour under dynamic spot pricing"
 },
 "session_cap": {
  "wall_clock_min": null,
  "idle_reap_s": null,
  "restart_required": false,
  "note": "a stopped VM stops the compute charge; the page states no automatic reap"
 },
 "persistent_storage": null,
 "network_out": null,
 "price_note": null,
 "free_tier_kind": "none",
 "burst_x": 1.0,
 "ps": {
  "gib": null,
  "price": "volumes and shared filesystems bill separately, running or stopped",
  "durability": "volume",
  "on_stop": true,
  "on_destroy": false,
  "retention": "while the volume exists"
 },
 "net": null
}

Unknown or withheld values (17)

$.chaining not-stated
$.egress.metered egress-unknown
$.egress.terms not-stated
$.evidence[0].proved process
$.extras.keep_rate_basis process
$.extras.price_note process
$.identity_cost.detail defaulted
$.identity_cost.kind identity-defaulted
$.interface interface-unknown
$.interface_detail withheld
$.limits_note not-stated
$.per_object_cap.notes process
$.per_object_cap.stated not-stated
$.reliability_note none-observed
$.retention.notes not-extracted
$.retention.stated not-stated
$.tooling not-stated

Nebius research grants Compute

fails: identityinstitutionno cardDwrite-gated

nebius.com/nebius-research-grants · GPU cloud · compute/nebius-research-grants · JSON · compare

Free tier	GPU cloud credits, Token Factory tokens and Tavily credits. The page states 'The exact amount is determined individually based on each proposal' and states no ceiling. Eligibility is a closed list of four roles, all academic: 'Postgraduate students conducting academic research', 'PhD students and postdoctoral researchers', 'Faculty members from accredited academic institutions', 'Researchers from accredited academic institutions or non-profit research organizations'. unit none; metered on wall-clock; one-time-credit
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	institution not published
Limits	write-gated the eligible-role list names only students, postdocs and staff of accredited institutions, so an unaffiliated applicant has no category to apply under. source the eligible-role list names only students, postdocs and staff of accredited institutions, so an unaffiliated applicant has no category to apply under.
Binding	write-gated the eligible-role list names only students, postdocs and staff of accredited institutions, so an unaffiliated applicant has no category to apply under.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	None usable: the award is denominated in the vendor's own credits and tokens, so it does not compose with another provider's allowance, and the identity gate precedes it.
Reliability	no reliability event recorded
Notes	Recorded because the amount is unbounded and the program is live, which makes it a magnet for the next session. The absence of a published ceiling is the trap: 'determined individually' means the record has no number to compare against any other entry, and the four-role list has no independent lane. A co-marketing obligation is attached ('being open to publicly acknowledging Nebius support'), which is a publishing condition rather than a compute one. Nothing here is a reason to reject it as a product; the door is the finding.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (2)

date	source	what it proved
2026-10-22	nebius.com/nebius-research-grants	the 'Who can apply' list names exactly four roles, every one tied to an accredited academic or non-profit institution; 'The exact amount is determined individually based on each proposal'; page 200
2026-10-22	nebius.com/nebius-research-grants	'We are now accepting applications for the 2026-2027 academic year', 'Nebius Academy will review applications throughout summer 2026', 'Successful applicants will be notified by email in late Q3 to early Q4 2026' - a seasonal cycle, not a rolling one

More fields (1)

{
 "wave2": {
  "class": "research-credit",
  "free_tier_kind": "one-time-credit",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": null,
   "gpu_count": null
  }
 }
}

Unknown or withheld values (10)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.detail process
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated