Best offer on record: Inference Providers free account: NO monthly credits (as of Oct 2026). 15 verified findings, checked 2026-10-22.
Offers
Inference Providers free account: NO monthly credits (as of Oct 2026) high confidence
Official pricing docs (read 2026-10-22): Free Users row in the included-credits table is 'None' — free accounts get no monthly compute credits. Free users can still access Inference Providers (18 providers, 200+ models: Baseten, Cerebras, Cohere, DeepInfra, Fal, Featherless, Fireworks, Groq, HF Inference, Novita, Nscale, OVHcloud, Public AI, Replicate, Scaleway, Together, WaveSpeed, Z.ai) via pay-as-you-go 'credits purchase required', or via a custom provider key. The PRO marketing page claims free-tier users get a 100k token/month inference usage cap vs 2M for PRO, but the pricing docs define no free credit and I could not find a published limits table — treat the 100k figure as marketing, not docs.
Eligibility
Any Hugging Face account (free, global, no .edu, no credit card to create the account or to browse; a credit card IS required before any pay-as-you-go usage, since extra usage requires purchasing credits via Stripe). No student or startup special tier.
How to claim
Sign up at huggingface.co, generate a fine-grained token with 'Make calls to Inference Providers' permission; add a credit card in Billing settings and purchase credits to exceed the free usage cap.
Expiry
Ongoing; no stated expiry
Gotchas
Trial-vs-free: there is no trial-credit offer on Inference Providers — the only ever-free allowance was $0.10/month for free accounts, discontinued 2026-10-07 (docs update). No credit card stated as required at sign-up, but one is required to buy pay-as-you-go credits. Monthly tokens cap (100k, unverified in docs) is a usage limit, not a free grant.
Custom Provider Key: $0 on Hugging Face (bring-your-own provider key) high confidence
Documented in the official pricing page (read 2026-10-22): you can set a custom provider key in HF settings (or pass it to the InferenceClient / OpenAI-compatible client) and HF routes requests for free — 'Hugging Face will not charge you for the call'; billing goes directly to the provider you already pay. This is the only genuinely card-free-on-HF way to use Inference Providers.
Eligibility
All account types incl. free. You need your own account/key with the provider; no region limit stated; no credit card needed on HF.
How to claim
Settings → Inference Providers → add custom provider key, or pass the key to InferenceClient; code and OpenAI-compatible SDK usage are unchanged.
Expiry
Ongoing
Gotchas
Included monthly credits do NOT apply to custom-provider-key usage. The provider's own free tier/trial terms now govern cost — this is a 'no markup' pass-through, not a free allowance. If the provider later requires its own payment method, the 'free' path ends there.
PRO plan: $9/mo subscription including $2.00/mo compute credits (paid, not free) high confidence
PRO ($9/month, credit card only, Stripe, no stated trial) includes $2.00/month of general-purpose compute credits usable on Inference Providers, Inference Endpoints, upgraded CPU/GPU Spaces hardware incl. ZeroGPU usage beyond the 5-min free daily quota, and Jobs. PRO also raises the inference usage cap to 2M tokens/mo (vs 100k free claim), 1 TB private storage (base price $12-18/TB/mo above), 10 ZeroGPU Spaces, 8× daily ZeroGPU quota (40 min RTX Pro 6000 Blackwell), dev-mode Spaces. No student discount, no free PRO tier, no student-specific enrollment found on huggingface.co/pro or /pricing today.
Eligibility
Any user globally; credit card required for the subscription. No .edu requirement, no regional restriction stated.
How to claim
Subscribe at https://huggingface.co/subscribe/pro; credits are credited monthly and applied automatically before pay-as-you-go billing.
Expiry
Renews monthly; cancel anytime in billing settings. Docs say credits are 'credited every month' but do not promise rollover of unused credits.
Gotchas
The $2 credit does not cover the $9 subscription. 'Pay Pro with credits' was reportedly possible in the past and now appears removed (HF forums, Feb 2026). Not a student/startup offer.
Team plan: $20/user/mo incl. $2/seat/mo shared credits (paid, not free) high confidence
Team orgs ($20/user/mo; self-serve, credit card or AWS/Google Cloud Marketplace billing) get $2.00/month compute credits per seat, shared across members, spendable on Inference Providers, Inference Endpoints, Spaces upgrades and Jobs. Org admins can set spending limits and disable providers; usage billable centrally via X-HF-Bill-To header or bill_to client parameter.
Eligibility
Orgs globally; card or cloud-marketplace billing. No startup/education discount published.
How to claim
Subscribe Team in org settings; members specify bill_to org (or X-HF-Bill-To header) to draw on the shared credit pool.
Gotchas
Credits only apply to HF-routed usage, not custom provider keys. Not free; included only for completeness of the 'credits' picture.
Six months free PRO for startups — third-party claims only, unverified official low confidence
Several affiliate/promo-code directories (SaaSTweaks 'verified May 2026', xraise.ai 'verified 2026-05-15', startup-perks.com 'audited Feb 2026') claim a 'Hugging Face for Startups' offer of ~6 months of free PRO plus Inference Endpoints credits, applied at huggingface.co/startups. But huggingface.co/startups is actually an org profile page ('startupsdao') with zero program content, and two independent checks of HF-owned pages found no program: guptadeepak (checked 2026-09-13: 'No startup program appears on any Hugging Face owned page') and CloudRoute's 2026-10-09 catalog, which excluded Hugging Face from its 62 verified startup programs because 'no official source confirmed a current program'.
Eligibility
Unverifiable — no published criteria on any HF page. Aggregators loosely describe 'early-stage AI startups', pre-seed to Series A, global; some mention 'oss' targeting.
How to claim
No official application path exists today. The claimed route (promo code at huggingface.co/startups) leads to an org page, not a program. Contact sales via /contact/sales if pursuing, or treat the offer as unconfirmed.
Gotchas
Affiliate-driven content; 'vendor-published criteria' cited by these sites are not on any huggingface.co page I could read today. Do not plan around this offer without written confirmation from HF.
Academia Hub organization tier — exists in docs, no public free offer low confidence
The Hub rate-limits docs (tiers table dated Sept '25) list an 'Academia Hub organization' tier with the same 5-minute rate limits as a Team org (3,000 API / 20,000 resolvers / 400 pages per 5 min), implying HF runs a dedicated academic org plan. However no huggingface.co page I read today (pricing, enterprise, pro, education) publishes an Academia plan, a student discount, an education license, a .edu-email program, or a free academic tier. huggingface.co/education is just an org profile ('High School' org, 15 followers, no program content). A community dataset notes the 'huggingface.co/community-grants' URL was 404 as of 2026-Q2 and no public grant application path existed; the /pricing page mentions 'community GPU grants for side projects' without a link.
Eligibility
Unknown/presumably universities & research institutions via contract or sales; no public eligibility, no .edu requirement published.
How to claim
No public claim path found; would be via /contact/sales. Treat as unavailable for self-serve students.
Expiry
n/a
Gotchas
This is an internal tier name surfaced in docs, not an offer. No amounts, no free access, no application form — do not cite it as an education license.
Free static Spaces hosting (unlimited) high confidence
Static Spaces (plain HTML, or built React/Svelte/Vue) are $0, free for everyone, unlimited in number, with no hardware charge and no card required. Default per-Space limits: 2 vCPU / 16 GB RAM / 50 GB non-persistent disk; CPU Basic (2 vCPU / 16 GB) has 'no hourly cost' but as of mid-2026 creating a compute Space (Gradio or Docker, incl. CPU Basic) requires a paid plan (PRO $9/mo personal; Team $20/Enterprise $50 per user for orgs). Static SDK Spaces never require a paid plan. All free for anonymous visitors too; no egress fee noted for Spaces.
Eligibility
Any account, free; no credit card, no age/region limit stated for static Spaces. No .edu requirement.
How to claim
Create a Space with the Static SDK at huggingface.co/spaces -> 'Create new Space', push your site; it deploys free forever. Zero setup cost, card only needed if you later add paid hardware.
Expiry
Indefinite while account exists; auto-rebuild on each push. Space goes to sleep when idle.
Gotchas
As of ~July 2026 free accounts can no longer create Gradio/Docker compute Spaces (incl. free CPU Basic) — static only; CPU Basic downgrade from ZeroGPU also requires PRO. Non-persistent 50GB disk, 16GB RAM ceiling. PRO is $9/mo card-only; free tier has no hourly charge but compute creation is gated.
ZeroGPU free GPU demos (host up to 2 Spaces + free usage quota) high confidence
ZeroGPU dynamically allocates NVIDIA RTX Pro 6000 Blackwell (half = 48GB VRAM 'large', full = 96GB 'xlarge', xlarge costs 2x quota). Free personal accounts 'in good standing' (verified email, account older than 30 days) can HOST up to 2 ZeroGPU Spaces at $0. Visitors get daily GPU quotas: unauthenticated 2 min, free account 5 min, PRO 40 min (8x), Team 40 min, Enterprise 60 min; quota resets exactly 24h after first usage. Default per-call runtime 60s; over-quota usage $1 per 10 min, but only PRO/Team/Enterprise can extend (pre-paid credits). No card required for the free tier.
Eligibility
Free personal accounts: email verified + account >30 days old; Gradio SDK only; PyTorch-first (transformers/diffusers best). No .edu or startup requirement; no card required.
How to claim
Create a Gradio Space with ZeroGPU hardware selected, decorate GPU code with @spaces.GPU, push; hosting is free for up to 2 such Spaces. Usage by any visitor is free within their daily 5-min (free account) quota.
Expiry
Indefinite; daily quota resets every 24h. PRO is $9/mo (card required) for 10 Spaces, 40 min quota and overage billing.
Gotchas
Free accounts cannot buy over-quota credits (paid tiers only). xlarge (96GB) burns 2x quota (45s task = 90s quota). After 2026 hardware migration some workloads auto-fell back to xlarge, doubling effective quota consumption (forum reports May 2026). Queue priority rises as your remaining daily quota falls. Gradio-only; Docker Spaces can't use ZeroGPU. Historically an undocumented run-count limit (~3 runs/day, staff-confirmed April 2026) may still apply on top of the 5-min time quota. A forum answer (July 2026) suggests new free accounts see ZeroGPU as the only creatable runtime — CPU Basic now PRO-gated.
Community GPU grants (in-kind, rolling) medium confidence
HF offers 'community GPU grants' for side projects: in-kind dedicated GPU hardware that ZeroGPU can't cover (non-PyTorch heavy init, models >96GB VRAM, always-on serving) or an extra Gradio Space beyond the free 2-Space cap without needing PRO. No dollar amount, duration or pool size published; decisions can take days; approved hardware attaches automatically (Static Spaces just flip SDK to gradio).
Eligibility
Public, open use cases — 'open research demo, hobbyist project, educational tool, institutional showcase' — that 'can't pay for the hardware'. No funding-cap or age limit; free account works; card not needed for the grant itself.
How to claim
1) Build your Space (ZeroGPU or Static). 2) Post a Community Tab discussion titled 'Apply for a GPU community grant: <Personal|Company|Academic> project' with a one-paragraph description + justification (default: 'Public open-source demo, can't cover the hardware cost'). 3) Wait for approval; hardware attaches automatically.
Expiry
Rolling, no published deadline; no published grant duration.
Gotchas
In-kind only, no cash; approval is discretionary and slow (days+); not a formal startup program; don't rely on a guaranteed quota.
Hugging Face for Classrooms (free education workspace) high confidence
Free classroom-organization offering: teachers set up a shared workspace where students upload datasets, models and ML demos 'for free', with version control and student access controls (admin/read/write), teaching materials and support. 'Everything is free' per the page; no numeric repo/storage/GPU caps published.
Eligibility
Any classroom — students and teachers; no .edu verification gate stated, no card required, no institution-type minimum.
How to claim
Create free at huggingface.co/classrooms ('Create a classroom for free', routed to /organizations/new?orgType=classroom); students join and host demos/datasets in the org workspace at $0.
Expiry
Indefinite while the classroom org exists.
Gotchas
Compute (GPU) is not part of the free classroom bundle — it's storage/hub hosting of repos, datasets, models and static demos; hardware upgrades are separate paid services. Limits not published, so heavy use could hit general free-tier storage limits.
Academia Hub (education license — PAID, free only at partner universities) high confidence
Institution-level education plan: from $10/seat/month with volume discounts; minimum 250 annual seats; includes 1TB private + 1TB public storage per seat, $2/month/seat of compute credits (Inference Providers, Inference Endpoints, upgraded Spaces hardware, Jobs), higher daily ZeroGPU priority, SSO/SCIM/SOC 2. Students get nothing for free as individuals — HF staff confirmed there is no student Pro plan (Oct 2025 forum). Partner universities (e.g., ESCP, 12,000+ students) negotiate free access for their community.
Eligibility
University IT/procurement must initiate from a university email; individual .edu email is NOT enough. Free only at partner institutions that have negotiated a deal.
How to claim
Contact HF from your university's IT/procurement team to negotiate; or at a partner university, your admin provisions your seat — then you get PRO-equivalent features + $2/mo compute credits free of personal cost.
Not free for most students; the $2/mo compute credits are for paid seats the institution funds; minimum 250 seats excludes most small departments (Team plan suggested instead).
Startup program: none first-party (third-party 'free PRO for startups' claims unverified) low confidence
No startup program exists on any Hugging Face owned page (checked Sep 2026 by a tracker; pricing/enterprise pages list only PRO/Team/Enterprise with no startup discount or application path). The well-known 2024 '$10M of GPUs' was a one-off commitment at ZeroGPU launch (its free quota persists, the $10M pool did not). Third-party perk marketplaces (creditforstartups, saasoffers, perq.sh) advertise 'up to $108 free PRO via partner perks' or '$1,000 in GPU credits' for HF — none of these could be verified against an HF-owned page; treat as unverified aggregator claims. The old '50% off Enterprise in year one' startup offer is no longer verifiable anywhere on HF's sites.
Eligibility
N/A for a first-party program; third-party perks claim pre-seed–Series A AI startups worldwide, referral from partner investor/accelerator, but terms aren't published by HF.
How to claim
There is nothing to claim directly. Startups should use: free static Spaces + 2 free ZeroGPU Spaces per team member, the community GPU grant (in-kind), and generic cloud startup programs (AWS Activate, Google/Microsoft for Startups) for compute instead.
Expiry
n/a
Gotchas
Beware unverified marketplace 'credits' (amounts unverifiable, may require equity/referral, may be expired promos). Meta x Hugging Face x Scaleway AI Startup Program (EU accelerator, 2nd cohort Sep 2024–Feb 2025) is now discontinued.
ZeroGPU free GPU inference (Spaces) high confidence
Shared NVIDIA RTX PRO 6000 Blackwell GPUs allocated on demand to Gradio Spaces. 'large' = half card, 48GB VRAM (1x quota); 'xlarge' = full card, 96GB VRAM (2x quota). Daily GPU quotas: unauthenticated 2 min; free account 5 min; PRO/Team 40 min; Enterprise 60 min; paid plans can extend via pre-paid credits at $1/10 min. Quota resets exactly 24h after first daily GPU use. No credit card required for the free tier; no stated regional limits. Free accounts host max 2 ZeroGPU Spaces (PRO: 10; orgs: 50). Gradio SDK only; @spaces.GPU decorator, default 60s function runtime (extendable via duration=...); no torch.compile (use PyTorch AOT, torch 2.8+).
Eligibility
Any free HF account in good standing: verified email address and account older than 30 days. No .edu required, no startup status, no credit card, no stated region limits. Using (visiting) existing ZeroGPU Spaces is free for everyone, even unauthenticated (2 min/day).
How to claim
Sign up at huggingface.co (free), verify email; create a Gradio Space and select ZeroGPU hardware in Space Settings (free accounts: up to 2 such Spaces, allowed only after account is 30+ days old); import spaces and decorate GPU functions with @spaces.GPU (default size large; size='xlarge' for 96GB at 2x quota cost); deploy and use within the 5 min/day free quota.
Expiry
Quota renews daily (resets 24h after first use). Hosting cap (2 ZeroGPU Spaces) persists while account stays in good standing; no sunset date published for the program.
Gotchas
Gradio-only (Docker/static cannot use ZeroGPU); xlarge doubles quota consumption and queue wait; free tier cannot buy quota extension ($1/10-min credits only for PRO/Team/Enterprise); model must be loaded to 'cuda' at module level (CUDA is emulated outside @spaces.GPU functions); remaining quota determines queue priority, so idle free users queue behind paid ones; as of 2025-2026 creating new CPU Gradio/Docker Spaces now requires a paid plan - the 2-Space ZeroGPU exception is the only free Gradio path.
Static Spaces + free Hub tier (free hosting, no GPU) high confidence
Static HTML Spaces are free for everyone on any plan. CPU Basic hardware (2 vCPU/16GB, no accelerator) is $0/hour. Free personal accounts get 100GB private storage (public storage 'best-effort' free), unlimited public models/datasets repos, and free CPU Spaces. No credit card needed. Inference Endpoints/Jobs (paid GPU) are NOT free - there is no free notebook/training offering.
Eligibility
Any free HF account, worldwide, no verification beyond the account itself; static Spaces need no email-verified/30-day status the way ZeroGPU hosting does.
How to claim
Create a free account, create a Space with SDK 'static HTML' (or CPU Basic Gradio/Docker if already on a paid plan); no payment setup required.
Expiry
Ongoing free tier; no expiry stated. Note the 2025-2026 change: new Gradio/Docker Spaces (even on free CPU Basic) now require PRO (personal) or Team/Enterprise (orgs) to create or duplicate - only Static Spaces remain free for free accounts.
Gotchas
The marketing page /classrooms still says demos are 'hosted for free', but the Spaces docs explicitly require a paid plan for new Gradio/Docker Spaces; duplicating a Gradio/Docker Space follows the same paid rule (ZeroGPU exception only). Free spaces sleep when idle; protected/private visibility requires PRO/Team/Enterprise. Free private storage capped at 100GB, overflow is billed.
Community GPU grants (free dedicated GPU hardware for Spaces) medium confidence
Case-by-case grants that cover Space hardware upgrade costs (dedicated GPU, or ZeroGPU slots beyond the free 2-Space cap) for public-facing community use cases: open research demos, hobbyist projects, educational tools, institutional showcases; company projects may apply ('Company project' option). No fixed dollar amount - the grant attaches the needed hardware to the Space. Official docs link to applying 'using the link in the Settings tab of your Space'. No credit card; no stated startup age/funding caps.
Eligibility
Qualifying public, non-commercial-leaning Space use cases where the user 'can't pay for the hardware' (open source, research, educational, institutional). Private/commercial/closed-source projects are steered to PRO instead. No .edu required for company grants.
How to claim
Build the Space (as ZeroGPU if you still have a free slot, else a free Static Space with ZeroGPU-ready code), then open a Community Tab discussion on the Space titled 'Apply for a GPU community grant: <Personal|Company|Academic> project' with one paragraph on what the app does and one on why it should run on free GPU; wait days for review; approved grants attach hardware automatically (Static holding Spaces must flip sdk to gradio). The entry point is also the 'apply for a community GPU grant' link in the Space Settings tab.
Expiry
Grant covers the attached hardware while the Space/project qualifies; no published renewal term. The old public application page (huggingface.co/community-grants) is reported dead as of 2026-Q2; the current channel is the per-Space Settings tab link / Community discussion.
Gotchas
Approval takes days and is discretionary ('see if yours makes the cut'); open public demos typically approved, commercial ones not; grant docs come mainly from the official huggingface/skills repo rather than a dedicated marketing page, so policy can shift without announcement; no guarantees on hardware size.
Bottom line as of 2026-10-22: Hugging Face has NO live free offer specifically for students or startups on the Inference Providers front. (1) STARTUP PROGRAM — sunset: the former 'HF for Startups' program (widely reported as $1,000/mo of partner cloud credits for up to 12 months plus ~50% off the Enterprise Hub in year 1; original announcement page now 404) no longer appears on any HF-owned page — independently confirmed Sept 2026 (guptadeepak) and Oct 2026 (CloudRoute excluded HF from its verified program catalog). Third-party '6 months free PRO' claims are affiliate promo-code pages, unverified against official pages; huggingface.co/startups is an org profile, not a program page. (2) FREE ACCOUNT CREDIT — sunset: free accounts historically received a small monthly Inference Providers credit ($0.10/mo, 'subject to change'; TechCrunch 2025-01-28: 'all HF users get a small quota of credits'); the official pricing docs were updated 2026-10-07 to 'None' for free users, and only PRO ($9/mo, $2/mo credits) and Team/Enterprise ($2/seat/mo) receive included credits. (3) EDUCATION — no education license or student offer exists on any HF page today; an 'Academia Hub' org tier appears only in the rate-limits docs (no pricing, no application path, not free). (4) TRIAL CREDITS — HF publishes no trial-credit offer for Inference Providers (no signup bonus, no 30-day trial of PRO; a 2026 forum thread indicates paying for PRO with credits was also removed). Only genuinely free routes: the free Hub/Spaces tier (CPU Basic + ZeroGPU, 5 min/day, no card), the $0 custom-provider-key routing path (BYO provider key; provider's own free tiers apply), and 'community GPU grants for side projects' mentioned without a live application URL. No regional restrictions found; global. Sources: [Hugging Face Inference Providers pricing (official, read today)](https://huggingface.co/docs/inference-providers/en/pricing) · [Hugging Face docs hub rate limits](https://huggingface.co/docs/hub/en/rate-limits) · [Hugging Face docs hub billing](https://huggingface.co/docs/hub/billing) · [Hugging Face PRO](https://huggingface.co/pro) · [Hugging Face pricing](https://huggingface.co/pricing) · [Hugging Face enterprise](https://huggingface.co/enterprise) · [guptadeepak: HF Startup Program discontinued (Sep 2026 check)](https://guptadeepak.com/startup-offers/programs/huggingface-startups) · [CloudRoute startup-credits changelog Oct 2026](https://www.cloudroutehq.com/startup-credits/changes) · [Free LLM Credits changelog (HF free credit ended Oct 2026)](https://freellmcredits.com/changes) · [PerkBook: HF for Startups 'No Longer Available'](https://perkbook.co/startup-programs/hugging-for-startups/) · [SaaSTweaks: 6 months free PRO claim (affiliate, unverified)](https://saastweaks.com/startup-credits/hugging-face-pro-credits) · [TechCrunch: Inference Providers launch, Jan 2025](https://techcrunch.com/2025/01/28/hugging-face-makes-it-easier-for-devs-to-run-ai-models-on-third-party-clouds/) · [HF forum: paying for PRO with credits no longer possible](https://discuss.huggingface.co/t/is-it-not-possible-to-pay-for-pro-with-credits-anymore/173037). Note: web-search budget was exhausted mid-task; all confidence labels above reflect pages I fetched directly today unless flagged as third-party/unverified."
2025-2026 material changes: (1) Mid-2026 (docs PR merged 2026-07-21; forum reports from 2026-07-09): creating compute Spaces — Gradio or Docker, including formerly-free CPU Basic — now requires a paid plan (PRO $9/mo personal, Team/Enterprise for orgs); free personal accounts were pushed to 'Static only, or up to 2 ZeroGPU Spaces'. This materially shrank the free tier that previously let anyone run a free CPU Gradio Space. (2) 2026-06-29: ZeroGPU hosting formalized a free tier — 'accounts in good standing' (verified email, >30 days old) host up to 2 ZeroGPU Spaces; PRO up to 10; paid orgs up to 50. Before this the free hosting cap/wording changed repeatedly. (3) December 2025: H100 / 8x H100 removed from the Spaces hardware lineup. (4) ZeroGPU hardware migrated from A100 (2024 launch) to NVIDIA RTX Pro 6000 Blackwell by 2026; post-migration some workloads silently fall back to xlarge, doubling quota consumption (forum reports May 2026). (5) No student plan has ever shipped; HF staff confirmed in Oct 2025 that Academia Hub is the only education path. (6) The $10M GPU pledge (May 2024, ZeroGPU launch, via A100s) was a one-off commitment — the durable free offer is the ZeroGPU daily quota, not a $ pool. (7) Payment gotcha: PRO is credit-card-only; India-issued cards are not yet supported (use AWS/Google Cloud Marketplace billing workaround). Free account and all free-tier usage require no card. No region limits on free usage found.
No free education license or startup program: Hugging Face's education offering is Academia Hub, a PAID institutional plan ($10/seat/month, min 250 annual seats, incl. $2/seat/month compute credits and higher ZeroGPU priority) — institutions like ESCP (Aug 2025) pay for it; there is no self-serve .edu free program and no dedicated free startup program (no application form, no funding-cap criteria found; community GPU grants are the closest equivalent and accept 'Company project' cases). No free ML training/notebook compute exists: HF Jobs and Inference Endpoints are on-demand paid, and the only free inference-side compute is ZeroGPU's 5 min/day plus CPU Basic $0/hr. Material 2025-2026 changes: (1) new Gradio/Docker Spaces now require a paid plan (PRO personal / Team-Enterprise org) — previously CPU Spaces were free for all; free accounts retain only Static Spaces + up to 2 ZeroGPU Gradio Spaces (verified in the spaces-overview doc, 2026-10-22). (2) ZeroGPU hardware is now shared NVIDIA RTX PRO 6000 Blackwell (48/96GB slices) replacing the earlier A100/A10G-based shared fleet; the original program (May 2024, ~$10M commitment, The Verge) survives in this form. (3) The former public community-grants application page (huggingface.co/community-grants) is reported 404 as of 2026-Q2; grants are now requested per-Space via the Settings tab link / Community discussion. (4) A May 2026 press item (CBW News) claims ~$5/month free GPU credits for all accounts; this does not appear on huggingface.co/pricing or the official docs and is unverified (low confidence). Free-account 100GB private storage, best-effort public storage, and 'protected' visibility gating to paid plans also confirmed 2026-10-22. WebSearch session budget was exhausted after 2 searches, so sunset/change claims rest on the official pages fetched today plus third-party changelogs (aifree.dev, perq.sh, The Verge, ESCP).
Critic corrections
Hugging Face Inference Providers: Contradicted by official docs. The Inference Providers pricing page (read Oct 2026) still gives Free users $0.10 of monthly credit (marked subject to change), PRO users $2.00 per month, and pay-as-you-go after that. The claim that free accounts get none is not supported. Oct 2026 forum threads report some new free accounts showing $0.00 balances, so availability looks inconsistent. Restate as reduced to $0.10/month, subject to change.
outdated: creating Gradio or Docker Spaces now needs a paid plan: A September 2026 secondary source says creating Gradio or Docker Spaces needs PRO ($9 per month) or Team, while free accounts can still host two ZeroGPU Spaces. Update the free-tier wording and verify on the Hugging Face pricing page before republishing.