Home / FAQ
FAQ: 23 questions about free AI APIs, free VPS & VPS deals
The questions below use English users’ actual wording, gathered from search, forum, and community demand research. Answers are answer-first, based on vendor official pages; where an answer relies on community sources rather than a vendor official page, that is stated. Check the official page before claiming.
Are free LLM APIs really free?
Some are genuinely free long-term; most are capped allowances, not unlimited usage. The 43 verified offers split into four shapes: 2 long-term free tiers, 18 one-time signup credits, 23 periodic resets (13 daily, 1 weekly, 9 monthly), and 2 with fixed end dates. “Free” always comes with conditions — an account, rate limits, an expiry, or all three. See the free-tier types comparison for the full grouping.
Do free AI APIs require a credit card?
Only 2 of the 43 verified offers list card binding as a requirement: Cerebras ($5 trial credit) and Vercel AI Gateway ($5/month). The other 41 don’t — use the “No card required” filter on the overview. But “no card” doesn’t mean “no requirements”: many offers need an account, a phone number, or real-name verification. One caveat: Cloudflare Workers AI notes that some newer models require a payment method first.
What is the best free LLM API with no credit card?
It depends on what you need. Six verified LLM APIs need no card: Gemini (Flash models, context up to 1M), Groq (speed, per-model limits), OpenRouter (:free models, one key for many providers), Cloudflare (10,000 Neurons/day), Zhipu GLM Flash (officially marked fully free), and SiliconFlow (¥14 credit + voucher, real-name verification required). Full comparison: best free LLM APIs with no credit card.
Is Claude Code free? Does the free Claude plan include Claude Code?
The tool is free; the model usage usually isn’t. The Claude Code command-line tool can be downloaded and installed for free, but calling models consumes API credits or a subscription. Per community-compiled Anthropic documentation, the Free plan does not include Claude Code — using it needs a paid subscription or API credits. Check Anthropic’s official page. The coding offers (Qoder, WorkBuddy / CodeBuddy, TRAE, Kiro student plan, Dify Cloud) are credits inside their own clients or clouds — not Claude Code credits.
Is the Gemini API free tier available for every Gemini model?
No. The verified listing describes the free tier as covering Gemini Flash and other models, at 5–30 RPM with per-model daily caps; the claim steps mention Gemini Flash / Flash-Lite. Which models are included, and their limits, are per the AI Studio quota panel. See the Gemini API free tier detail page and check Google’s official docs.
Is there a free web search API?
Yes — four verified ones. Tavily (1,000 API credits/month), Exa ($10/month + a $10 signup gift), Firecrawl (1,000 credits/month, scraping), and You.com (100 searches/day via MCP, no account or key). None requires a card. Comparison: free search & scraping APIs compared.
Is Pinecone free?
Pinecone has a free Inference Starter allowance: 5M embedding tokens per model per month + 500 reranks/month, issued monthly and separately per model, account required. The verified offer covers inference (embedding + rerank) allowances; other Pinecone usage is per the official pricing page.
Is there a completely free and unlimited AI API?
No. Every verified offer has a rate limit, a credit cap, or an expiry. The closest in the verified list: Bilibili Index-Translate (no registration, no cap stated on the official site — but translation-only, 150 languages, not a general chat model) and Zhipu GLM Flash (officially marked fully free, with per-account concurrency). “Unlimited” claims elsewhere are not in the verified data — see the four allowance shapes.
What do RPM, RPD, and TPM mean? When do quotas reset?
RPM = requests per minute; RPD = requests per day; TPM = tokens per minute. Some vendors also limit TPD (tokens per day). All four axes apply at once — hitting any one returns a 429. Verified examples: Groq is “per model 30 RPM / 1,000 requests per day / 200K tokens per day”; OpenRouter free models are “20 RPM / 50 requests per day”; Gemini free tier is “5–30 RPM / per-model daily caps”. Reset times differ per vendor: Cloudflare Workers AI resets at 00:00 UTC (08:00 Beijing time); Huawei Cloud CodeArts at 24:00 (UTC+8); AMD at 00:00 Beijing time; IFM at 00:00 UTC. See each offer’s “reset” field.
What happens when I hit a limit or a trial ends? Will I be charged?
It depends on the vendor — three patterns exist. Stops: Tencent Cloud TokenHub notes usage stops without auto-charging — postpaid billing must be enabled to continue. Switches to pay-as-you-go: Alibaba Cloud Bailian switches by default — turn on “stop when exhausted”; Qiniu also switches to pay-as-you-go when used up. Overage billing: You.com bills overage at $5 per 1,000 searches. Where what happens is not stated, confirm on the official page before claiming: is a card required, is there a “stop when exhausted” switch, does anything auto-renew? Never assume “free allowance” means no charge.
I got a 429 error. Should I wait a minute or until tomorrow?
Check which limit you hit. Minute-level limits (RPM/TPM): wait a bit and retry. Day-level limits (RPD/TPD): wait until the reset time. Some vendors send a Retry-After header (suggested wait in seconds) — follow it. Don’t retry blindly: OpenRouter explicitly notes that failed requests also count toward the daily cap, so retrying burns the daily allowance faster. Day-level waits depend on each vendor’s reset time (see the previous question). Longer-term relief: lower concurrency, exponential backoff, and spreading requests across vendors (without violating vendor terms).
Is the free web chat the same as the free API allowance?
No — they are two separate systems. Free web/app chat and open-platform API allowances don’t share quotas, may list different models, and bill independently. Web free doesn’t mean API free, and vice versa. For example, community-compiled Q&A notes that DeepSeek’s free web version doesn’t mean the API is free, and Gemini’s app/web version and API free tier are separate. Judge only by the platform’s API allowance wording. Every offer is tagged by access type: API, client app, or web/cloud.
Why can’t some free models plug straight into Codex CLI or Claude Code?
The most common cause is a protocol mismatch. Different tools require different protocols: Claude Code requires the Anthropic protocol; Codex CLI requires the Responses API; Cline, OpenCode, and similar tools mostly use OpenAI-compatible formats. A free key that only offers OpenAI Chat Completions format can’t plug directly into a tool requiring another protocol. The second common cause is a model that doesn’t support function calling, which coding tools rely on. The third is a wrong Base URL — for example, adding /chat/completions twice, which returns a 404. Configuration details are per each tool’s official docs; listings mark which offers are OpenAI-compatible (23 of 43).
How do I fill in the Base URL? Do I need to add /chat/completions?
There is no single answer — it depends on the tool. Some tools want the Base URL ending at /v1 and append /chat/completions themselves; others want the full request URL. If both sides add /chat/completions, requests go to /chat/completions/chat/completions and return 404 — on a 404, check for a doubled path first. Only Bilibili Index-Translate states a Base URL in its claim steps (index-translate.bilibili.com/v1, model Index-Translate-35B-A3B); every other offer’s Base URL and model ID are unstated — check the official page.
Will my data be used to train models? Can I use free tiers commercially?
Policies differ by vendor — no blanket answer. It is noted only when a vendor states it explicitly: Google AI Studio (Gemini API free tier) notes that free-tier data may be used to improve Google products. For other offers, no data/training policy or commercial-use terms are stated — check each vendor’s official terms, especially before commercial use.
Can I get a free VPS without a credit card?
Only one verified offer explicitly requires no card: Azure for Students ($100 credit / 12 months) — but it is students-only, with school-email verification. AWS’s official FAQ says most new customers need no payment method to register, though identity verification may be required in some cases. Oracle, Google Cloud, and Kamatera all require a card. Alibaba Cloud, Tencent Cloud, and InstaBox don’t state card requirements on their official pages. Full table: Free VPS list.
Is Oracle Cloud Always Free really free forever?
There is no fixed end date while the account stays active — but “forever” comes with rules. The A1 allowance is 2 OCPU / 12 GB since June 2026 (the 4 OCPU / 24 GB in old tutorials is outdated; the PAYG 4-core-free claim is disputed). Idle instances may be reclaimed under the 7-day rule; capacity errors are common; there is no SLA; and the $300 trial credit is separate (30 days). Details: Oracle Always Free detail page.
Why do I get “Out of host capacity” for VM.Standard.A1.Flex?
Your home region temporarily has no Always Free capacity for that shape. The official guidance: switch availability domain, wait and retry, or upgrade to PAYG for more compute (Always Free resources still cost nothing after upgrading — only usage above the free limits is billed). The official FAQ says adding capacity may take several days. The AMD-based E2.1.Micro is an alternative shape.
Will Oracle reclaim my idle Always Free instance?
Yes, it can be reclaimed. The official docs’ test: over 7 days, CPU 95th-percentile utilization below 20%, network utilization below 20%, and memory utilization below 20% (A1 shapes only) counts as idle. Accounts idle for more than 30 days may also be treated as abandoned and suspended or terminated. Don’t fake real usage with load scripts.
Is AWS Free Tier still 12 months free?
No. New accounts moved to the $200 credit / 6-month model (from July 15, 2025, per second-hand sources’ date; the official FAQ describes only the credit model): $100 on signup + up to $100 more for exploring foundational services, with no fixed free EC2 VM, and it cannot be extended. When the free plan expires, the account is suspended and data is kept 90 days, then permanently deleted if not upgraded. Note: the official FAQ gives two wordings for credit expiry — 6 months and 12 months — so go by the official page and billing page. Most new customers need no payment method to register.
Are VPS trial credits real? How do I know the credit actually arrived?
They are real offers with real conditions — and one is disputed. Vultr’s match requires a deposit first and offsets 50/50 (12 months); Hetzner’s €50 has the clearest rules but only lasts the current billing month; Akamai’s $100 is solid but the trial length in days is unstated and GPU plans are excluded; DigitalOcean’s $200 is disputed — community answers say direct signup only gets $5/90 days. Always check the credited amount in your account right after signup, before deciding whether to use it. Comparison: trial credits compared.
Is the renewal price the same as the first-year price?
It depends on the vendor. RackNerd: yes — the official FAQ promises lifetime same-rate renewal (“lifetime recurring, renew at the same rate”). HostEONS: not stated on the plan pages — confirm on the checkout page and don’t assume. DigitalOcean: list price with a monthly cap, no intro discount. Trial credits have no renewal at all, only expiry rules. See the Cheap VPS list and the cheap vs free comparison.
Annual or monthly — which costs less over 3 years?
With stated renewal prices, annual wins; with unstated renewal, the 3-year total is unknowable. Calculated from stated prices: RackNerd 1 GB at $21.99/year × 3 = $65.97 (same-rate renewal stated in the official FAQ); DigitalOcean Basic 512MiB at the $4.00/month cap × 36 = $144.00 (list price, monthly cap). HostEONS is excluded from the math — its renewal price is unstated, so no 3-year total can be computed. Monthly billing buys flexibility (cancel any time); annual billing buys the lower total where renewal is stated.
Disclaimer: answers follow the vendor official page content; allowances can change or end at any time — check the official page before claiming.