Free AI API without a card: the full 2026 map.

Quick answer: who gives keys without a card.

The vendor landscape splits three ways. Genuinely card-free: Google AI Studio issues a Gemini key with no billing account, GitHub Models rides on a GitHub login, Groq and Mistral run free tiers on registration, and OpenRouter serves its rotating :free model variants on a bare account. Prepaid-first: OpenAI, Anthropic, and xAI all expect a funded balance before sustained API usage, and the automatic trial credits people remember have ended or turned unreliable. In between: Chinese vendors vary, with Z.ai serving older Flash models free while the current families bill per token everywhere. The catch pattern is consistent: card-free tiers are prototype-sized by design, metered in requests per minute and per day, and several pay for themselves with your data. The table maps the routes this site has covered in detail; each links out to its own page below.

Why cards gate free tiers at all.

A card requirement is a fraud filter, not a revenue play. Free inference gets farmed: scripted account creation, key resellers, and abuse traffic will drain any uncapped free tier within days, and a card on file is the classic way to make each account cost something to mint. Vendors that skip the card compensate somewhere else: aggressive rate caps, phone verification, account-age checks, or terms that let them train on your traffic. That explains the two honest shapes a no-card offer can take. Either it is small and capped, like the studio tiers and free variants above, or it is a metered credit that simply starts the meter at a positive balance, like the route here. Anything advertising unlimited free inference with no card and no catch is describing someone else's stolen keys, and those die fast and log everything. Reading a no-card offer therefore takes one question: what does the vendor get? If the answer is "a shot at converting you to paid usage", the offer is sustainable and safe to build an evaluation on. If the answer is unclear, your prompts are probably the payment.

What the no-card starting credit here covers.

The starting credit is issued without a card and spends on any catalog id: the Claude family, the GPT family, Gemini, DeepSeek, GLM, Kimi, Qwen, MiniMax, and Grok, all behind one OpenAI-compatible endpoint and one key. There is no free-model list and no degraded free lane; the credit meters the same models at the same catalog rates as paid traffic, and nothing you send is training payment for the tokens. The honest limits, stated plainly: it is a starting credit, so it ends; it is metered, so heavy models spend it faster than value models; and afterwards access is pay-as-you-go top-ups, which also work without a stored card and carry no subscription. An idle month costs nothing. What makes a multi-family credit worth more than the same value on a single vendor is the comparison it funds. The vendor-by-vendor free tiers above each answer "does this vendor's model work for me" in isolation; one credit across the catalog answers "which model should carry this workload", which is the question that actually decides your steady-state bill.

curl https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $APISROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Three subject lines for a beta invite email."}]
  }'

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

The whole free family, one page each.

Each linked page applies the same standard as this one: what the vendor actually offers, dated and hedged where reports conflict, and where the no-card credit here fits among the honest routes.

• Claude: no vendor free tier at all; the chatbot is free, the API is prepaid. Covered in the free Claude API guide and the direct-answer page linked below.
• GPT: trial credits ended; the real vendor mechanism is opt-in data-sharing tokens. Covered in the free GPT API guide.
• Gemini: the most generous genuine free tier of the big labs, via Google AI Studio. Covered in the Gemini free guides.
• DeepSeek, GLM, Kimi, Qwen: value families where free routes include open weights and, for GLM, official free Flash models. Each has its own page.
• Grok: no standing free tier, launch promos come and go. Covered in the free Grok API guide.
• Everything at once: the free LLM API options roundup compares the cross-vendor mechanisms, including AI Horde and local models.

When no-card free stops being the right constraint.

A card aversion usually encodes one of three real concerns, and each has a cleaner answer than hunting ever-smaller free tiers. If the concern is surprise charges: prepaid top-ups cannot overdraft, which is true of the route here and of most vendor platforms now. If the concern is subscriptions: pay-as-you-go with no monthly fee removes the treadmill, and an idle month costs nothing. If the concern is simply not owning a usable card: top-up rails that take non-card payment solve it directly, and the starting credit means even that decision waits until the evaluation is done. The signal to move past free is the same as on every page in this family: the first time a cap costs more time than the metered price would have cost money. At value-family rates, the workloads that outgrow free tiers usually land at single-digit dollars per month, which is a small price for never seeing a 429 mid-demo again.

FAQ