DeepInfra

deepinfra.com · Compute, Inference · 3 records · JSON

DeepInfra Sandboxes ComputepaidcardcardDretention-clamped
DeepInfra InferencepaidcardcardD
DeepInfra DeepStart Computefails: identityinstitutionno cardDretention-clamped

DeepInfra Sandboxes Compute

paidcardcardDretention-clamped

deepinfra.com/blog/sandboxes · microVM (Kata on QEMU/KVM) sandbox · compute/deepinfra-sandboxes · JSON · compare

Free tier	no free tier: 'the nano plan starts at just $0.054/hour, billed by the second'; no credit or trial is offered for sandboxes on the pages read unit none; metered on wall-clock
Price	none stated
Per-object cap	not applicable: the unit is a sandbox, not an object. Each command defaults to a 60-second timeout with a 30-minute hard cap.
Retention	Stopping keeps /workspace and resets everything else; 'you've got up to 7 days to start() it again before it's permanently deleted'. Only /workspace survives a stop, and failed sandboxes lose /workspace. (7 days)
Egress	Not published. The overview page and the blog post state the compute and lifecycle meters and say nothing about a traffic allowance or a per-GB egress rate.
Identity	card A DeepInfra account is required and sandboxes draw on DeepInfra billing; the catalog endpoint refuses an unauthenticated caller, so plan shapes and prices are visible only to a signed-in account. Card types, any authorisation hold, and whether a debit or prepaid card is accepted were not established.
Limits	retention-clamped auto-stop after 24 continuous running hours, a default 1-hour idle timeout that cannot be disabled, and 7 days of stopped retention before deletion source volume-quota 25 non-stopped sandboxes per account; 'shared-fleet capacity can temporarily prevent creation' source Billing runs while a sandbox is creating, starting, running or stopping, so 'per-second' includes boot and the disk save on stop. The plan shapes and rates are behind an authenticated catalog: GET /v1/sandboxes/catalog returned {"detail":"User is not authorized to access this resource"} to an unauthenticated request, and the docs say explicitly not to hardcode the numbers.
Binding	retention-clamped The 24-hour auto-stop is the first hard wall: no configuration on the pages read lifts it, so the machine cannot be an always-on monitor or VPN node even before the price matters.
Interface	rest, cli Python SDK ('Sandbox.create(plan="medium")') and an HTTP API, with a dashboard at deepinfra.com/dash/sandboxes.
Tooling	The vendor's Python SDK and REST API; commands run inside the microVM and files move only through /workspace.
Composes with	USD 0.054 an hour for the nano plan is USD 39.42 a month if never stopped, and a sandbox auto-stops after 24 hours of continuous running, so it is not an always-on machine.
Reliability	Two documented loss paths: a failed sandbox loses /workspace, and everything outside /workspace is discarded on stop. Shared-fleet capacity can refuse creation.
Notes	Arithmetic: USD 0.054 an hour × 730 = USD 39.42 a month if the auto-stop did not exist; × 300 h = USD 16.20. The auto-stop makes the 730-hour figure hypothetical, and it is recorded as the price of the meter rather than the price of a machine. What the nano shape actually contains is unverified because the catalog is authenticated. The dossier already carries a DeepInfra record for its DeepStart programme; this is a different product from the same vendor and does not supersede it.
Evidence rung	documented The nano rate, the per-second meter and the whole lifecycle are quoted from DeepInfra's own blog and docs, read today. The rate-to-shape mapping is not documented publicly: the vendor directs readers to an authenticated catalog, so the shape fields stay null rather than guessed. What would raise it: a signed-in read of /v1/sandboxes/catalog, which would give the exact vCPU/RAM/disk of the nano plan.
Checked	2026-10-25 (live); stale on 2026-11-24
Parsed numbers	`$.extras.m10h_usd` price USD16.2 / month `$.extras.m24h_usd` price USD39.42 / month

Evidence (4)

date	source	what it proved
2026-10-24	deepinfra.com/blog/sandboxes	quoted: 'the `nano` plan starts at just $0.054/hour, billed by the second' and 'Leave one running and it auto-stops after 24 hours'
2026-10-24	deepinfra.com/blog/sandboxes	retention: 'Stopping just pauses it: /workspace sticks around, everything else resets to a clean image. From there you've got up to 7 days to start() it again before it's permanently deleted.'
2026-10-24	docs.deepinfra.com/sandboxes/overview	billing: 'Billing is per-second with no minimum, and only runs while a sandbox is creating, starting, running, or stopping - a stopped sandbox costs nothing'
2026-10-24	docs.deepinfra.com/sandboxes/overview	no public shape/rate table: 'For current plan specs (vCPU, RAM, disk) and hourly pricing, don't hardcode numbers - check the live catalog'

More fields (11)

{
 "shape": {
  "vcpu": null,
  "ram_gib": null,
  "disk_gib": null,
  "gpu_class": null,
  "gpu_count": null
 },
 "billing_model": "hourly",
 "term": "per-second metering, no commitment",
 "m10h_usd": 16.2,
 "m24h_usd": 39.42,
 "price_arithmetic": "0.054/hr × 730 h = USD 39.42; × 300 h = USD 16.20. Both assume the sandbox is never auto-stopped, which the vendor's 24-hour rule makes impossible in practice.",
 "free_tier_kind": "none",
 "network_out": {
  "allow_outbound": true,
  "allow_inbound_listen": null,
  "ports": "not published; the docs describe command execution and file transfer, not an inbound listener",
  "static_ip": null,
  "ipv6": null,
  "per_session_bytes": null
 },
 "session_cap": {
  "wall_clock_min": 1440,
  "idle_reap_s": 3600,
  "restart_required": true,
  "note": "auto-stop at 24 h continuous; 1 h default idle timeout; 7-day stopped retention"
 },
 "aggregator_gross_usd_month": null,
 "aggregator_note": "sandboxprices.com has a DeepInfra Sandboxes row marked unverified with no total, for the same reason recorded here: the shape-to-rate mapping is behind the authenticated catalog."
}

Unknown or withheld values (1)

$.egress.metered egress-unknown

DeepInfra Inference

paidcardcardD

api.deepinfra.com/v1/openai · inference endpoint, keyed free tier · inference/deepinfra · JSON · compare

Free tier	no signup credit: 'You have to add a card or pre-pay or you won't be able to use our services', quoted. The only free balance is a promotional USD 20 for viewers of the Two Minute Papers channel, claimed through a deal link at login unit requests; metered on request-count
Price	none stated
Per-object cap	not applicable: an inference endpoint stores nothing for the caller
Retention	not applicable
Egress	the response is the payload; no separate egress term is published
Identity	card the pricing page states a card or a prepaid balance is required before any use
Binding	none
Interface	rest
Tooling	any OpenAI-compatible client
Composes with	pairs with an agent after a signup; the scarce link is the account and the unread allowance is the gap
Reliability	no reliability event recorded; no key was available
Notes	The weakest record in this topic and it says so. It exists to stop a name becoming a rediscovery. The pricing page states the opposite in one sentence, and the only free balance is a channel promotion. Paid-only is the correct verdict and documented is the correct rung.
Evidence rung	documented neither the page nor the endpoint was reached with evidence; the row is carried so the next session does not re-discover the name. verified_at is a typed midnight and not a machine-read instant, so it is an upper bound on when the record was confirmed rather than the instant; the evidence dates carry the real reading day. The 401 shape was re-verified at 2026-10-22T04:42:43Z; the quota remains unread, so the rung stays unverified.
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (2)

date	source	what it proved
2026-10-22	deepinfra.com/pricing	'You have to add a card or pre-pay or you won't be able to use our services. Pre-paid accounts draw down their balance as they go', quoted
2026-10-22	deepinfra.com/papers	'we're thrilled to offer you $20 in free credits to explore DeepInfra's AI inference cloud', claimed through a login deal link, quoted

More fields (1)

{
 "wave3": {
  "verdict_change": "qualifies-with-caveat -> paid-only on 2026-10-22"
 }
}

Unknown or withheld values (1)

$.retention.stated not-applicable

Corrections

2026-10-22 Compute deepinfra requires a card

DeepInfra DeepStart Compute

fails: identityinstitutionno cardDretention-clamped

deepinfra.com/deepstart · inference API · compute/deepinfra-deepstart · JSON · compare

Free tier	50,000,000,000 free inference tokens for an approved startup, priced at DeepSeek V4 Flash rates and expiring one year after approval. At the DeepSeek-V4-Flash-0731 rate of USD 0.06 per million input tokens the grant is worth USD 3,000 of input, or USD 9,000 at the USD 0.18 output rate. unit none; metered on wall-clock; grant
Price	none stated
Per-object cap	not applicable
Retention	the funding gate fails before the tokens do: 'Have raised between 250K and 10M USD' and 'Founded in the last 2 years' are company facts, and the credits also expire one year after approval.
Egress	not stated
Identity	institution not published
Limits	retention-clamped the funding gate fails before the tokens do: 'Have raised between 250K and 10M USD' and 'Founded in the last 2 years' are company facts, and the credits also expire one year after approval. source the funding gate fails before the tokens do: 'Have raised between 250K and 10M USD' and 'Founded in the last 2 years' are company facts, and the credits also expire one year after approval.
Binding	retention-clamped the funding gate fails before the tokens do: 'Have raised between 250K and 10M USD' and 'Founded in the last 2 years' are company facts, and the credits also expire one year after approval.
Interface	rest inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	none for a user: a grant needs a funded company, so it cannot compose with anything a single person holds.
Reliability	no reliability event recorded
Notes	50,000,000,000 tokens / 1,000,000 = 50,000 units of a million tokens; 50,000 x USD 0.06 = USD 3,000 input-equivalent; 50,000 x USD 0.18 = USD 9,000 output-equivalent. A third-party list circulating on 2026-10-22 states the grant is 1,000,000,000 tokens, which the vendor page contradicts by a factor of 50; the vendor figure is recorded. This is the largest per-entity inference credit found and it is closed to a person without a funded company. It is a separate programme from the signup record above: the same provider carries both a closed grant and a card-gated API.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (2)

date	source	what it proved
2026-10-22	deepinfra.com/deepstart	'DeepStart ~50,000,000,000 free tokens' with 'To qualify your startup needs to meet the following: Have raised between 250K and 10M USD; Founded in the last 2 years; at DeepSeek V4 Flash prices; Credits expire 1 year after approval', quoted
2026-10-22	deepinfra.com/pricing	DeepSeek-V4-Flash-0731 at USD 0.06 per 1M input and USD 0.18 per 1M output tokens, used for the conversion, quoted

More fields (1)

{
 "wave2": {
  "class": "startup-grant",
  "free_tier_kind": "grant",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": null,
   "gpu_count": null
  }
 }
}

Unknown or withheld values (8)

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