free-llm-api-hub
A continuously-verified dataset of free-tier & trial-credit LLM APIs for developers. Every entry dated, sourced to the provider's own docs, and machine-readable. No hype, no dead links.
Automatically checked · GitHub checked 2026-11-14

Overview
A continuously-verified dataset of free LLM & AI-model APIs you can build on.
Free LLM APIs plus adjacent model APIs — image, speech, embeddings, rerank and OCR. Free tiers, trial credits and no-cost quotas, every entry dated, sourced, and machine-readable. Every entry is dated and sourced to the provider's own docs; what we could not confirm is marked. Just what's actually free, and the fine print that bites.
🔎 Interactive explorer · 📊 Dataset · 🧪 How we verify · ➕ Add a provider
⭐ Useful? Star the repo so other builders find it. · 🛠 Want to help? Pick a good first issue or a help wanted issue; follow its named files and check. New here? Follow the first-contribution steps.
Why this exists
Most "free LLM API" lists are a snapshot someone took once and never touched again. Rate limits get cut, docs URLs move, "free forever" quietly becomes "free for 30 days" — and the list keeps recommending it, confidently, for another year.
This project treats the list as a maintained open dataset, not a blog post:
- Every entry is dated and sourced. No
Last verifieddate and a link to the provider's own docs? It doesn't ship as verified. - Freshness is measured, not claimed. The build grades the oldest re-verification against the 90-day SLA. The explorer calculates current ages in the browser; the README badge shows the last committed refresh.
- Links are checked, not scheduled. Every change runs the integrity gate, and each re-verification pass opens the provider's own docs — a dead source link is the earliest signal a provider changed something.
- Uncertainty is labelled, not hidden. Entries we couldn't independently confirm are marked ⚠️ and say exactly what's unconfirmed, instead of being dressed up as fact.
- The data is the source of truth.
data/providers.jsonis validated against a schema; the marked data sections of this README, the badge and the provider site are generated from canonical inputs and checked for drift. Authored prose remains subject to review; see the input map.
If you're picking an API to prototype on this afternoon, you want the fine print more than the marketing. That's the whole point of this repo.
[!WARNING] Independent, community-maintained list — not affiliated with, endorsed by, or sponsored by any provider below. Free-tier terms change without notice. Always confirm against the provider's own docs (linked in every row) before you rely on anything here. Entries marked ⚠️ are sourced from community tracking and not yet independently re-confirmed — treat them as indicative.
What's covered
Primarily free LLM (text) APIs — plus the adjacent model APIs a builder reaches for next: image generation, speech (STT/TTS), embeddings, rerank and OCR. Every category is held to the same verification bar. Breakdown is generated from the data:
| Category | Providers | Examples |
|---|---|---|
| Text / LLM | 44 | Google Gemini API, Groq, OpenRouter |
| Speech (STT / TTS) | 26 | Google Gemini API, Groq, Cloudflare Workers AI |
| Embeddings | 13 | Google Gemini API, Cloudflare Workers AI, Cohere |
| Image generation | 11 | Cloudflare Workers AI, HuggingFace, OVHcloud AI Endpoints |
| Vision | 11 | Google Gemini API, OpenRouter, Z.ai |
| OCR / documents | 8 | OCR.space, LlamaParse, Nanonets |
| Rerank | 6 | Cohere, Jina AI, Mixedbread |
Filter any category live in the interactive explorer or the multimodal collection.
Browse by API requirement
The guides and collections compare dated provider rows, limits and catches.
| Requirement | Compare |
|---|---|
| No payment card | No-card tiers and trials |
| No signup | Account requirements explained — no phone and no card does not mean no account |
| OpenAI SDK | OpenAI-compatible endpoints — feature support can differ |
| Embeddings / RAG | Embedding APIs |
| OCR / documents | OCR and document parsing |
| Voice | Speech-to-text and text-to-speech |
| Images | Image generation and editing — check which operations each tier includes |
Video generation is not currently a tracked modality in the dataset. There is no verified video comparison here; image or vision support does not establish video-generation access.
TL;DR — pick by what you actually need
Each pick links to its full verified profile on the live site.
| I want… | Start with | Why |
|---|---|---|
| The smartest model, free | Google Gemini | The only genuinely frontier-class model with a real free tier here — not just open weights |
| The fastest inference | Groq or SambaNova | Purpose-built inference chips — far faster than typical GPU-served APIs |
| The most free volume/day | Cloudflare Workers AI (10k Neurons) or OpenRouter (1k req/day) | Highest ceilings for a side project with real traffic |
| No card and no phone | OpenRouter or Google Gemini | Groq, SiliconFlow and NVIDIA all gate signup behind phone verification |
| Open weights (Llama, DeepSeek, Qwen, GLM) | OpenRouter or Cloudflare Workers AI | Widest open-model selection on an ongoing free tier |
| Permanently free, no trial clock | Z.ai (GLM), or SiliconFlow if you can pass China real-name verification | Several models priced at $0 indefinitely, not just for a trial window |
| An OpenAI-compatible endpoint | Groq, OpenRouter, Cloudflare Workers AI | Point the OpenAI SDK at a new base_url and you're done |
| EU / data-sovereignty hosting | OVHcloud or Scaleway | French/EU providers; OVHcloud even has an anonymous, no-account tier |
| Free embeddings & rerank | Jina AI or Cohere | 10M free tokens (Jina, OpenAI-compatible) or 1,000 calls/mo (Cohere) |
| Free speech-to-text / TTS | Deepgram or AssemblyAI | $200 / $50 in no-card credit for Whisper-class STT and TTS |
| A bigger one-time credit | Deepgram ($200, speech) or AI21 Labs ($10 for 3 months, LLMs) | Large one-time credits that start without a card |
| Something safe to ship commercially | Cloudflare Workers AI or Groq | Don't restrict the free tier to personal/eval use, the way Cohere and NVIDIA do |
Starting points, not guarantees — read the full profile before you build on it.
Browse by need
Focused, always-current collections — each is generated from the dataset and has a live web page too.
- Free LLM APIs with no credit card (55) — start without a payment method · live page ↗
- Free LLM APIs with no phone verification (23) — no SMS/phone verification · live page ↗
- Free LLM APIs for commercial use (27) — safe to ship, not eval-only · live page ↗
- OpenAI-compatible free LLM APIs (35) — drop-in OpenAI SDK swap · live page ↗
- Permanently free LLM APIs (5) — $0 models, no trial clock · live page ↗
- Free multimodal LLM APIs (52) — vision, audio, embeddings · live page ↗
- Free LLM APIs that say they do not train on your prompts (19) — says it does not train on your prompts (some only without opt-in) · live page ↗
- Free LLM APIs whose free tier renews (28) — free quotas or credits that reset · live page ↗
- Free LLM APIs with a free signup credit (33) — one-time signup credits, not renewing allowances · live page ↗
Every provider also has its own page with the full details and a copy-ready quickstart — e.g. Groq, Deepgram, Jina AI.
Contents
- What's covered — the categories, by the numbers
- Browse by need — curated collections by constraint
- The best free LLM APIs — our top 10 by score
- Notably NOT free — so this list doesn't waste your time
- How verification works — the trust engine
- How this project is maintained — agents under human oversight
- Use the data — dataset, exports, badge
- Contributing · Project docs
The best free LLM APIs
Our top 10 by score — computed from the 65 verified providers every time the site is built, never hand-ordered and never padded to ten. The three best form the podium. Money adds nothing to the score. Every row links to its full verified profile: free tier, rate limits and the catch are checked against the provider's own docs. The podium and the method are on the /best page ↗ and in how the score works ↗.
💳 no card · 📵 no phone · 📱 phone required · 🏢 commercial OK · 🔬 eval only · 🔌 OpenAI-compatible
| # | Provider | Score | What's free | The catch | Verified |
|---|---|---|---|---|---|
| 1 | OpenRouter 💳 no card · 📵 no phone · 🏢 commercial OK · 🔌 OpenAI-compat | 67.2 5 of 6 inputs | A rotating set of models with a :free suffix, single API across many providers; availability fluctuates | Terms prohibit reselling API access or developing a competing service; per-model terms still apply. Upstream providers may train on inputs/outputs. Separate settings for free and paid models can exclude providers that train; router-level opt-in logging and safety/legal retention exceptions still apply (official terms, privacy and provider-logging docs read 2026-11-14). | ✅ 2026-11-14 |
| 2 | Groq 💳 no card · 📱 phone · 🏢 commercial OK · 🔌 OpenAI-compat | 64.3 5 of 6 inputs | Open-weight models (GPT-OSS, Qwen) plus Whisper, no credit card required | Limits apply at the organization level, not per API key. Phone verification required at signup | ✅ 2026-11-14 |
| 3 | SambaNova Cloud 🏢 commercial OK · 🔌 OpenAI-compat | 58.8 4 of 6 inputs | Rate-limited free tier (applies when no payment method is linked) across the models listed in the Free Tier table | Free Tier applies when no payment method is linked to the account; SambaCloud ToS grants a commercial license (no evaluation-only clause). The previously-listed "$5 / 3 months" trial could not be re-confirmed on official pages (2026-07-30). card_required is unconfirmed because SambaNova's own pages disagree (read 2026-10-08): the rate-limits doc (https://docs.sambanova.ai/docs/en/models/rate-limits) says "Free Tier: Applied when there is no payment method linked with your account", while the plans page (https://cloud.sambanova.ai/plans) says "Add a payment method and purchase credits to run your first requests". Settling it needs a real signup, which has not been done. The rate-limits doc (read 2026-10-09) also lists two preview models on the Free Tier at the same 20 RPM / 20 RPD / 200,000 TPD (DeepSeek-V3.2, gemma-4-31B-it), described there as having limited capacity and removable at short notice; they are not in models_free. | ✅ 2026-10-09 |
| 4 | Google Gemini API (AI Studio) 💳 no card · 📵 no phone · 🏢 commercial OK · 🔌 OpenAI-compat | 57.5 4 of 6 inputs | Gemini 2.5 Flash, 2.5 Flash-Lite, 2.5 Pro (limited), embeddings, TTS models | Free-tier prompts/outputs may be used by Google to improve its products outside the UK/CH/EEA/EU. Since the 2026-03-23 terms, only Paid Services may serve API clients to end users in the EEA/CH/UK. Gemini 2.5 Pro and 2.5 Flash are deprecated; the pricing page announces automatic routing to Gemini 3.8 Flash and 3.5 Flash-Lite respectively, billed at their rates, from 2026-12-01 | ✅ 2026-11-14 |
| 5 | Cloudflare Workers AI 💳 no card · 📵 no phone · 🏢 commercial OK · 🔌 OpenAI-compat | 52.0 4 of 6 inputs | 10,000 Neurons/day, all account plans | Resets daily at 00:00 UTC; overage on a Workers Paid plan bills at $0.011/1,000 Neurons. Some models require a paid billing method: Workers Paid or prepaid AI Gateway credits; these are outside no-card free access | ✅ 2026-11-14 |
| 6 | Z.ai (Zhipu AI / GLM) 💳 no card · 📵 no phone · 🏢 commercial OK · 🔌 OpenAI-compat | 52.0 4 of 6 inputs | GLM-4.5-Flash, GLM-4.7-Flash (text), and GLM-4.6V-Flash (vision) are officially listed as $0 cost (input, cached input, and output) on a permanent basis | Terms of Use prohibit using Z.ai to "develop, train, or enhance any algorithms, models, or technologies that directly or indirectly compete with us" — otherwise general use, including commercial, isn't restricted | ✅ 2026-10-08 |
| 7 | ModelScope (API-Inference) 💳 no card · 🔬 eval only · 🔌 OpenAI-compat | 51.4 4 of 6 inputs | Free API-Inference calls paid with daily 'Magicubes': 200/day for signing in + 50/day after linking an Alibaba Cloud account (do not roll over); calls cost ~0.5 / 1 / 2 Magicubes for lightweight / standard / flagship models | Alibaba's model hub — a different product from Alibaba Model Studio. Requires a ModelScope account bound to an Alibaba Cloud account with real-name (ID) verification — a practical barrier for non-China users. Explicitly non-commercial ("for developers to experience"). No card. | ✅ 2026-10-08 |
| 8 | Cohere 💳 no card · 📵 no phone · 🔬 eval only · 🔌 OpenAI-compat | 46.1 5 of 6 inputs | Trial (evaluation) API keys covering chat, embed and rerank | Free evaluation key; production use requires upgrading to a production key. Cohere privacy policy says trial inputs and outputs may be used for research and product improvement; no trial-specific retention duration is confirmed | ✅ 2026-11-14 |
That's the shortlist. The full dataset — all 67 providers, every ongoing tier and one-time credit, filterable and machine-readable — lives in data/providers.json and the interactive explorer ↗.
Notably NOT free
Worth saying plainly. As of the last verification pass, OpenAI, Anthropic and xAI do not offer an ongoing free API tier. Several providers people assume are free — Together AI, DeepInfra, Perplexity's API, Replicate, Featherless AI — currently require a card or prepayment before any API use, per their own docs. Some have handed out small one-time trial credits at various points, but that's changed repeatedly; check each provider's billing page before assuming anything.
For genuinely free access to strong models, your best bets here are Gemini (frontier-class) and the free open-weight models on Groq, OpenRouter, Cloudflare and Z.ai (SiliconFlow's free models are on its China platform and need real-name verification).
Retired or removed after re-verification: Mistral (Free mode no longer includes API access; API keys require billing activation and pay-as-you-go, checked 2026-11-14), GitHub Models (fully retired by GitHub on 2026-07-30 — playground, catalog and inference API shut down for all customers), Cerebras (now listed as a trial credit: the ongoing free tier became a payment-method-gated $5/30-day trial), Inference.net (the free tier is gateway/observability only, not free model tokens), and Hyperbolic (not included pending a complete inclusion review; the current public-access check supersedes the earlier deposit finding).
How verification works
Free-tier terms move fast, and most lists go stale silently. This one is built to surface drift instead of hiding it. Full details in docs/methodology.md; the short version:
- Every verified entry carries a
last_verifieddate and a link to the provider's own docs. No date + primary source → it ships as ⚠️ unverified, not as fact. - A local re-verification pass (docs/update-playbook.md) re-checks each provider's own docs on a rolling basis — a dead or changed source is an early warning that a provider changed something.
- The freshness badge is computed from the data, not written by hand: it is graded on the oldest verification in the list, straight from
providers.json. Green while every verified entry is at most 60 days old and at least 70% of rows are verified (see the freshness rules), amber once any entry is due for re-verification, red once any entry breaches the 90-day SLA. One forgotten row is enough to move it on rebuild. The README badge reads the last committedbadge-freshness.json; it can age between committed refreshes even though Pages deploys daily. - The dataset is schema-validated in CI. A verified entry that's missing its date or source link fails the build — the honesty rule is enforced by machine, not by good intentions.
- Reporting a stale entry takes under a minute via a structured form that asks for the provider, what changed and a source link.
What "verified" covers and where its limits are: docs/methodology.md · what earns a spot on the list: docs/inclusion-criteria.md.
How this project is maintained
The maintainers are @pacocartones and @LeonMAG. AI agents assist under human oversight with repository checks, human-authored data updates and code. Provider re-verification dossiers are human-review material only; transfer the reviewed edit into a fresh worktree before agent-assisted regeneration, as described in the playbook. Every change still goes through a pull request, a human maintainer reviews it and is responsible for every merge, and every data change carries a link to the provider's own page and the date it was checked. Commit messages carry no AI signature or trailer; this section is where we say it.
Use the data
This is meant to be consumed by machines as much as by humans.
data/providers.json— canonical dataset, validated againstdata/schema.json. Every field explained in docs/comparison-dimensions.md.- Static JSON API — the public JSON dataset and the versioned JSON API are generated from the canonical dataset for programmatic use.
# Every ongoing free tier that needs neither a card nor a phone number:
curl -s https://raw.githubusercontent.com/pacocartones/free-llm-api-hub/main/data/providers.json \
| jq -r '.providers[]
| select(.category=="ongoing" and .card_required==false and .phone_required==false)
| .name'
Pin a snapshot for reproducible builds. main moves; depend on an immutable tag instead:
curl -s https://raw.githubusercontent.com/pacocartones/free-llm-api-hub/v2.11.0/data/providers.json # v2.11.0 was the latest on 2026-10-17; see Releases
Tags track the dataset version in data/providers.json (see CHANGELOG.md) — pin vX.Y.Z and bump deliberately. The live JSON API always serves the latest main.
Vouch for the data from your own README — embed the live freshness badge:
[](https://github.com/pacocartones/free-llm-api-hub)
It renders the age of the oldest verified entry at the last committed badge refresh. It does not recompute with each view; use the live explorer for freshness calculated in the browser.
Embed a live widget on any site — a compact, always-current list of the top verified free APIs:
<div id="flh-widget" data-count="6" data-modality="text"></div>
<script src="https://freellmapihub.com/widget.js" async></script>
Self-contained (inline styles, no CSS conflicts). All attributes are optional: data-count (1-20, default 6), data-modality (text, audio, embeddings, image, vision, ocr, rerank), data-category (ongoing or trial; default both), data-sort (an explorer sort key - recommended (default), name, category, free_tier, notes, verified).
Follow changes — updates page or the RSS feed.
Contributing
Found an outdated limit, a dead link, or a provider that belongs here? You'll keep this useful for everyone.
- Your first PR: choose a task, coordinate and run its checks.
- Fastest report: the structured issue form — provider, what changed, a source link.
- For a provider-data correction: edit
data/providers.json, runnpm run build && npm test, and commit the generated files with the data. Never hand-edit the tables — they're generated.
Full guidelines, including what counts as an acceptable source: CONTRIBUTING.md.
AI-assisted contributions are welcome
Use AI tools if they help you — we do too. We ask for good judgement: understand and test what you send, and check every fact at its source. See AI_POLICY.md.
🙋 Contributions wanted right now
Provider facts need checking, and the docs and site need care too. Three ways to help:
- Pick a good first issue. These include provider facts, docs and code; each issue names the files and the check. For a provider fact, including a docs source check, cite the official page and the day you read it (UTC). For code or other docs, follow the issue's evidence and acceptance criteria. A
good first issuealso labelleddatais not claimed; for other issues, comment/takebefore starting. Ready for a larger task? Follow the path from good first issue to help wanted. - Pick a provider from the umbrella issue confirm
phone_required. - Do the same for confirm
commercial_ok, reading the provider's terms.
For a provider fact, use the provider's own docs and a real date; if you cannot confirm it, leave the data unchanged and tell us what you read. /release frees an issue you claimed. We read every pull request and reply as soon as we can; see CONTRIBUTING.md.
October: help us verify, in the open, which AI APIs are really free. Contributions are welcome all year; Hacktoberfest 2026 no longer rewards pull requests, so none is promised here. One sourced fact beats ten cosmetic edits; see CONTRIBUTING.md.
Contributors
Thanks to everyone who has verified an entry, fixed a link, or improved the project:
- Jhansi Oruganti — data: confirm Novita AI free tier is not commercial-use allowed (PR #27)
- Victoria Odalo — docs: verify Clarifai commercial_ok status (PR #112)
- MikeGatsby — data: verify Datalab hosted-API commercial-use terms (no explicit statement found) (PR #117)
- bcabreraike-cmyk — docs: explain paced re-verification batches (PR #169)
- Swarnabha Nandi — fix: show clear button when sort is changed (PR #183)
- Manan Bharti — test: check serializer ORDER against the provider schema (PR #201)
- Piyush — data: verify Datalab phone requirement and free-tier rate limit (PR #198)
Project docs
| Doc | What it covers |
|---|---|
| Methodology | How each entry is verified; what "verified" does and doesn't mean |
| Update playbook | The weekly routine that keeps the badge green |
| Inclusion criteria | What earns a spot — and what gets rejected |
| Comparison dimensions | Every field and flag in the dataset, defined |
| Self-hosting on free compute | Adjacent: free GPU/compute when no hosted API fits |
| Credit programs (apply to get) | Adjacent: startup & student/research credit programs |
| Roadmap | Where this is going next |
| AI policy | AI-assisted contributions are welcome; what we ask in return |
| Changelog | What changed, when |
| Governance | How decisions get made |
| Security · Code of Conduct | Reporting & community norms |
License
MIT — free to reuse, fork, and adapt, including the dataset. A link back is appreciated but not required.