As of 2026-09-27, 24 free LLM API offers we track need no credit card at signup, by their recorded access and signup method. Read each row's caveats: our filter cannot, and a caveat can qualify the answer.
| Offer | Provider | Status | What it offers | Expiry | Verified | Source | Caveats |
|---|---|---|---|---|---|---|---|
| AI Gateway free tier | Vercel | available | A standing monthly free credit on AI Gateway for every Vercel team account, giving access to a subset of the model catalog (browsable at vercel.com/ai-gateway/models?freeTier=true) through one OpenAI… | no stated expiry | 2026-09-08 | source | THE AMOUNT IS NOT PUBLISHED FIRST-PARTY. Vercel's pricing page confirms a free tier and calls it "the monthly free credit" but states no figure; third parties widely report $5/month, which this row deliberately does not… |
| AI21 Jamba free trial | AI21 | available | $10 in credits for 7 days on the foundation-model API, covering Jamba Mini and Jamba Large. No credit card required to start. | no stated expiry | 2026-09-16 | source | A 7-DAY TRIAL, NOT A STANDING FREE TIER — the clock starts at signup and the credits expire with it, so this is unsuitable for anything durable. Verbatim from the pricing page: "Free Trial — $10 credits for 7 days. No c… |
| BazaarLink free gateway tier | BazaarLink | available | Free tier on an OpenAI-compatible gateway: 10 requests/minute and 50 requests/day, no credit card and no waitlist (“Sign up and get a free key in under 60 seconds. No credit card, no waitlist.”). Acc… | no stated expiry | 2026-09-08 | source | SMALL, AND SMALLER THAN AGGREGATORS CLAIM — third-party listings report 130 req/day and imply the whole catalog is free; the provider's own page states 50/day and shows exactly ONE model carrying a free quota (Qwen3.7 F… |
| Fireworks $1 starter credit | Fireworks AI | available | One-time $1 serverless inference credit granted at signup; access to serverless models while it lasts | no stated expiry | 2026-09-08 | source | Account is suspended once the credit is spent unless a payment method is added — that much is first-party. UNSOURCED FIGURE REMOVED 2026-09-08: this row previously stated “600 req/min” at confidence “verified”. That num… |
| Gemini API free tier (AI Studio) | available | Ongoing free tier (not a trial). Free of charge on Gemini 3.8 / 3.7 / 3.6 / 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 3 Flash Preview, 3.1 Flash TTS Preview, the 3.5 audio models (Transcribe, Transc… | no stated expiry | 2026-09-16 | source | free-tier prompts/outputs ARE used to train Google products; Pro models paid-only since Apr 2026; some region restrictions. CORRECTION 2026-09-16, WRONG IN BOTH DIRECTIONS. (1) This row listed Gemini 2.0 Flash as fr… | |
| GLM Flash models ($0) | Z.ai (Zhipu) | available | GLM-4.7-Flash and GLM-4.5-Flash (text) and GLM-4.6V-Flash (vision) priced $0 for input, cached input, and output — ongoing, OpenAI-compatible endpoint | no stated expiry | 2026-09-08 | source | PRC-based provider — assume training use / PRC data rules for sensitive data; ~1 concurrent request reported |
| GroqCloud Free Plan | Groq | available | Persistent free tier, no card. The rate-limits page embeds its figures as two tier-labelled arrays (freeRows / devRows); these are the freeRows values, read 2026-09-27. Free-plan models: openai/gpt-o… | no stated expiry | 2026-09-27 | source | THREE MODELS LEFT THE FREE TIER between 2026-09-08 and 2026-09-27: qwen/qwen3.6-27b (30 RPM / 1K req-day) and groq/compound and compound-mini (30 RPM / 250 req-day / 70K TPM) are absent from freeRows on 2026-09-27; the… |
| Hyperbolic Basic tier + $1 credit | Hyperbolic | available | Basic tier is free and rate-limited at 60 requests/minute; the provider's own table gives its requirement as simply “Free account”. Pro is 600/min on a $5+ deposit, Enterprise unlimited. All tiers ad… | no stated expiry | 2026-09-08 | source | PROVENANCE CORRECTION 2026-09-08 (third of three; see afr-0004 and afr-0013): this row carried access: "phone-verification" at confidence: verified, and said in free_what that the $1 credit was granted "on phone verific… |
| Inference Providers monthly credits | Hugging Face | available | $0.10/month included credits on free accounts, $2.00/month on PRO, and $2.00 per seat for Team or Enterprise organizations (pooled across members), usable across 200+ models on all integrated provide… | no stated expiry | 2026-09-08 | source | $0.10/month is tiny (~1-2M tokens of cheap models); only applies to HF-routed requests — bringing your own provider key means the credits do not apply and the provider bills you directly. PRO/Team/Enterprise credits are… |
| La Plateforme Free mode | Mistral | available | Free mode is the default on new accounts: the provider's own wording is that it “lets you create API keys and use included monthly usage within the limits shown on the Limits page”. Mistral publishes… | no stated expiry | 2026-09-08 | source | PROVENANCE CORRECTION 2026-09-08: this row carried access: "phone-verification" at confidence: verified, and its own source_url does not support it. A full read of that page today returns ZERO occurrences of “phone”; al… |
| Ling 3.0 Flash Fin and Sante ($0) | inclusionAI (via Novita AI and Vercel AI Gateway) | available | Ling 3.0 Flash Fin (finance) and Ling 3.0 Flash Sante (health) priced $0 for both input and output on Novita, 256K context; Vercel AI Gateway independently lists inclusionai/ling-3.0-flash-fin at $0… | limited-time, no date | 2026-09-08 | source | DO NOT CONFUSE WITH THE PAID SIBLING: plain “Ling 3.0 Flash” is charged on both gateways ($0.06/M in, $0.18/M out on Novita); only the -fin and -sante variants are free. These are domain-specialised models (finance and… |
| Model Studio new-user free quota | Alibaba Cloud | available | Typically 1,000,000 tokens per eligible model (qwen-max, Qwen3-Coder-Plus, etc. each get their own quota, not transferable between models), valid 90 days from the later of activation, model release o… | no stated expiry | 2026-08-26 | source | International (Singapore) endpoint only; enable 'Free Quota Only' mode to prevent auto-billing after exhaustion; 90-day expiry |
| NIM / build.nvidia.com developer credits | NVIDIA | available | 1,000 inference credits on free Developer Program signup (1 credit ~ 1 request; up to 5,000 more on request), ~40 req/min, 80+ hosted models via OpenAI-compatible API | no stated expiry | 2026-09-08 | source | PROVENANCE CORRECTION 2026-09-08 (second of three; see afr-0004 and afr-0016): this row carried access: "phone-verification" at confidence: verified and its source_url does not support it. That URL is a launch BLOG POST… |
| Novita new-user trial voucher | Novita AI | available | New-user trial voucher for Novita's model APIs — Novita's own quickstart states a voucher “with some credit” and gives no amount | no stated expiry | 2026-09-08 | source | PROVENANCE FIX 2026-09-08: source_url was previously a third-party aggregator (mintlify.com/cheahjs/free-llm-api-resources/providers/trial/novita), which violates this catalog's absolute first-party provenance rule; rep… |
| OpenCode Zen free models | OpenCode | available | Six models served at no cost through one OpenAI-compatible endpoint (https://opencode.ai/zen/v1/chat/completions), listed as "Free" for input, output and cached read in the published price table: Big… | no stated expiry | 2026-09-16 | source | A ROTATING SET, NOT AN ENTITLEMENT — treat the model list as a dated snapshot, the same way afr-0010 treats OpenRouter's :free variants. The free models are offered during feedback-collection periods and the table says… |
| OpenRouter :free model variants | OpenRouter | available | Rotating set of :free model variants — 17 of 458 models carried a :free suffix at $0 prompt and completion when read from openrouter.ai/api/v1/models on 2026-09-27: NVIDIA Nemotron 3 (nano-omni-30b r… | no stated expiry | 2026-09-27 | source | ROSTER MOVED 2026-09-16 → 2026-09-27: 20 of 443 became 17 of 458. Gone: InclusionAI Ling 3.0 Flash VL, Nex-AGI N2.5 mini and pro, Z.ai GLM-5.2. New: Qwen 3.8 27B. The limits page (read via a text-rendering route; it is… |
| OVHcloud AI Endpoints anonymous tier | OVHcloud | available | Anonymous, unauthenticated inference against the AI Endpoints model catalogue at 2 requests per minute, per IP and per model — no API key, no account, no signup of any kind. Verbatim: "Anonymous: 2 r… | no stated expiry | 2026-09-16 | source | 2 req/min is a trickle — this is for evaluation and fallback, not workload. Exceeding it returns 429; an API access key raises it to 400 req/min per PCI project per model. RATE-LIMITED PER IP, WHICH IS THE TRAP FOR AGEN… |
| Ramp Router | Ramp | available | Free routing through 2026 on an OpenAI/Anthropic-compatible endpoint fronting 26+ models (Claude, GPT, Gemini, Kimi, DeepSeek, Qwen, GLM, Grok); you pay provider list price for the tokens themselves… | no stated expiry | 2026-08-23 | source | launched 2026-08-20, so the terms are new and unproven; free routing is time-boxed — “free through 2026” — and the $26 credit is “subject to offer terms” that the landing page does not state; US-only (“individual develo… |
| SambaNova Cloud Free Tier + $5 credit | SambaNova | available | No-card Free Tier: 20 req/min but only 20 requests/DAY, 200K tokens/day (DeepSeek-V3.1, Meta-Llama-3.3-70B-Instruct, gpt-oss-120b); plus $5 signup credit | no stated expiry | 2026-08-28 | source | 20 requests/day makes the no-card tier nearly unusable for agents; credit expiry unclear (30d-3mo reported). Re-read 2026-08-28: every figure above still matches the provider's rate-limit table unchanged. The no-card co… |
| Space Bunny Alpha (stealth preview) | OpenRouter (anonymous third-party provider) | available | $0 prompt and $0 completion tokens on stealth/space-bunny-alpha during a limited-time anonymous preview: a model with adjustable reasoning, strong coding, and native multimodal input (text, image… | limited-time, no date | 2026-09-27 | source | A stealth preview: the provider is anonymous and states no end date. The previous stealth preview on this row's pattern (Ox Alpha, afr-0044) ran about a week and was then revealed as a paid model. Prompts and completion… |
| Ternary Bonsai 27B free serverless model | Together AI | available | Prism ML's Ternary Bonsai 27B served on Together AI's serverless infrastructure at $0.00/1M tokens for both input and output, with a 262,144-token context window. Reached through Together's ordinary… | limited-time, no date | 2026-08-28 | source | Zero price is confirmed on two first-party pages (the model page and together.ai/pricing) and the model is listed "Free / Free" on docs.together.ai/docs/serverless-models. NOT confirmed no-card: none of those three page… |
| Together AI signup credit | Together AI | available | Small one-time signup credit (third parties report $1-$25; amount not published). Separately, Together again serves at least one serverless model at $0.00/1M tokens - see afr-0055. | no stated expiry | 2026-08-28 | source | Treat the signup credit as a demo credit, not a free tier; its amount is still unpublished, which is why confidence stays "reported". CORRECTION 2026-08-28: this row previously asserted "formerly-free model endpoints ha… |
| Trial (evaluation) API keys | Cohere | available | Free trial keys, ongoing: 1,000 API calls/month total; Chat 20 req/min, Embed 2,000 inputs/min (image embed 5 inputs/min), Rerank 10 req/min | no stated expiry | 2026-08-27 | source | evaluation only - trial keys are 'not permitted to be used for production or commercial purposes' per cohere.com/pricing; that restriction is no longer stated on the rate-limits page at source_url, so read both. 1,000 c… |
| Workers AI free daily neurons | Cloudflare | available | 10,000 Neurons/day (~$0.11 of compute) at no charge, resetting 00:00 UTC — the same allocation on both the Workers Free and Workers Paid plans. Covers most of the catalog (Llama, DeepSeek distills, G… | no stated expiry | 2026-09-08 | source | 10K neurons is only ~10-30K tokens/day on large LLMs; mostly smaller/quantized open models. NOT ALL MODELS ARE INCLUDED — this row claimed “on all models” until 2026-09-08, and that is no longer true. Seven models are d… |
Eligibility isn't recorded as a field. Rows are included on how you sign up, not on who may.
For comparison, 3 available LLM API offers in our catalog do ask for a payment card at signup:
| Offer | Provider | Status | What it offers | Expiry | Verified | Source | Caveats |
|---|---|---|---|---|---|---|---|
| Cerebras Inference free tier | Cerebras | available | Free Trial: $5 in credits granted at signup, expiring 30 days later; while credits last, 5 req/min, 30K tokens/min, 1M tokens/hour, 1M tokens/day; access to all Cerebras-powered models | no stated expiry | 2026-09-08 | source | NOT a renewing free tier — Cerebras' docs state there is “no-cost tier that renews automatically” and require you to “add a verified payment method” to use the trial; credits expire 30 days after they are granted, and w… |
| Complimentary daily tokens (data-sharing) | OpenAI | available | Up to 1M tokens/day shared across the large-model group (gpt-5.6-sol, gpt-5.5, gpt-5.4, gpt-5.2, gpt-5.1, gpt-5.1-codex, gpt-5-codex, gpt-5, gpt-5-chat-latest, gpt-4.1, gpt-4o, o3, o1, o1-preview; 25… | no stated expiry | 2026-09-05 | source | NOT free-from-zero: requires a paid org at usage Tier 1+, and eligibility is per-organization - it shows on the org's data-sharing settings page; owner-level opt-in only; all shared traffic is used for training. Correct… |
| OpenRouter — $10 credits unlock 1000 free requests/day | OpenRouter | available | Purchasing at least 10 credits (~$10, one-off) raises the daily cap on :free model variants from 50 to 1000 requests/day. The per-minute limit is 20 RPM either way. The credits are yours to spend on… | no stated expiry | 2026-08-24 | source | Not free — it requires a real $10 purchase, which is why it is deal_type spend_unlock and not a free row; the underlying 50/day free tier (afr-0010) needs no spend. OpenRouter's docs state the threshold as "at least 10… |
Does a provider have a free tier? A–Z · Machine-readable: ai-deals.json