# Start an agent with no card and stay inside the free limits

> Start an agent with no card and stay inside the free limits: 1 companies fully do it, 15 with a limit, 44 carry the tag. Every listing in Model APIs & inference with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/inference.free-start/
- Markdown: https://www.anchorterminal.com/niches/inference.free-start/index.md (~2,700 tokens)
- Slim: https://www.anchorterminal.com/niches/inference.free-start/index.min.md (~530 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/inference.free-start/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

One company's docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 1 | 15 | 44 | 4 | 4 |

## What counts as done

Twenty agent turns of about 5K tokens each complete in a row without a 429 that cites a daily or free-tier quota, and the error names the quota when one is hit.

How agents ask for it: which API can my agent use for free; I'm on the free tier and get 429 after a few prompts; chain the free models so something always answers

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Cloudflare Workers AI (Cloudflare, Inc.)**: 10,000 neurons a day are free on the Workers Free plan, a 5K-token Gemma 4 26B turn costs a few neurons, and the 429 for a spent allocation carries code 3036 naming it; eight frontier models need Workers Paid and return 403 code 5035.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Claude API (Anthropic, PBC)**: New users receive free credits, but the pricing page does not say whether a card is needed, and new organisations start in an Evaluation tier with limits below the published Start tier.
- **Antseed (Antseed Foundation)**: A buyer with no deposits is served by free sellers only, yet automatic routing can rank a paid seller first and fail with 402, and a model with no free seller fails with 502 model_not_found.
- **BlockRun.AI (BlockRun, Inc.)**: Free models need no card and 429 FREE_TIER_RATE_LIMITED names the limit, but the cap is 30 requests a minute and 300 an hour per IP, which an agent making several calls a turn can exhaust inside twenty turns.
- **Cloudflare AI Gateway (Cloudflare, Inc.)**: Gateway functions are free on every plan, but models cost prepaid credits or your own provider key; only Workers AI's daily free neurons reach through it.
- **Cohere Chat API (Command models) (Cohere Inc.)**: Trial keys are free with no card at 20 chat requests a minute and 1,000 calls a month, barred from commercial use; the 429 body names the trial key limit, and an agent making several calls a turn can pass 20 a minute.
- **Fireworks AI Inference (Fireworks.ai, Inc.)**: A $1 starter credit before adding a payment method (pricing notes); the rate-limit page ties ceilings to spending tier, so the free start is a few calls.
- **Gemini Developer API (Google LLC)**: A free tier needs no card and 429 RESOURCE_EXHAUSTED names the quota, but acc-16 hit it after a few prompts, acc-17 saw paid Tier 1 projects cite the free-tier quota at limit 0, and acc-18 saw quota values that did not match the page.
- **GroqCloud (Groq LLC)**: Free with no card at 30 requests a minute, 1,000 a day and 8K tokens a minute on gpt-oss-120b, with the 429 carrying retry-after and headers naming the day's requests and the minute's tokens; several 5K-token calls a turn exceed the token limit.
- **Jais 3 (Mohamed bin Zayed University of Artificial Intelligence (Institute of Foundation Models), with Inception42 (a G42 company))**: Apache-2.0 weights cost nothing and need no card, but the smallest useful tool-calling run needs the owner's GPUs; 0.9B to 375B sizes are offered.
- **LiteLLM (Berrie AI Incorporated)**: Free to self-host under MIT with no account, but inference bills on the operator's own provider keys, so the free start is the software, not the tokens.
- **Nemotron 3.5 Super VL (NVIDIA Corporation)**: Free for Developer Program members with no card, but the FAQ calls the catalogue a trial whose request rate varies by model and concurrent users, a thread asks to raise 40 RPM, and no 429 body is documented.
- **OpenRouter (OpenRouter, Inc.)**: Free models at 20 requests a minute and 50 a day until $10 of credit is bought (then 1,000 a day), with free_model_daily_requests on GET /api/v1/key; acc-01 reports every free model returning 429 under a burst.
- **Router.com (Ramp Router) (Ramp Business Corporation)**: Free credit per the listing, but sign-up needs a US business email in a browser and no rate limits are published; 402 and 401 limit codes name the spend cap when one is hit.
- **Beam 1.0 (Reflection AI)**: A plan-set daily token allowance with no card for some organisations, but the amount isn't published, some must verify a card, and 429 daily_allowance_exhausted names the quota.
- **SambaCloud (SambaNova Systems, Inc.)**: Free tier with no payment method, but 20 requests a minute, 20 a day and 200,000 tokens a day per model, with headers showing what remains; twenty multi-call turns need rotating across models.
- **SWE-2 Flash (Cognition AI, Inc.)**: Free in Devin Desktop and CLI until 4 December 2026 18:00 UTC and SWE-2 Flash Medium is on the Free plan; High and Max need a paid plan, and plan quotas aren't published as numbers.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **MiniMax API (Nanonoble Pte. Ltd.)**: The pages read give no free LLM allowance and the pricing overview wasn't opened; the rate-limit page mentions a free tier only to exclude speech-to-text from it.
- **Novita AI (Novita AI)**: Two models are listed free per the pricing notes, but the rate-limit page is drawn by script and no free-tier quota or 429 body was readable.
- **OpenAI API (OpenAI OpCo, LLC)**: The Free tier row shows an allowed-geography requirement and a $100 monthly usage limit but no per-model limits or card requirement; which models it reaches wasn't readable.
- **SiliconFlow (SiliconFlow Labs Pte. Ltd.)**: $1 of credit per the listing; whether a card is needed and the free-tier RPM weren't readable because the rate-limits page returns 404.
- **Simplismart (Verute Technologies Private Limited)**: Free credits 'may' be granted per the listing; no rate limits or free-tier quota are published.
- **StepFun API (STEPFUN TECHNOLOGY (HK) LIMITED)**: The host platform.stepfun.ai closed the connection on every attempt (curl exit 56, no HTTP status) from this session, so no page was read.
- **Tencent Cloud TokenHub (Tencent Cloud Computing (Beijing) Co., Ltd. (腾讯云计算（北京）有限责任公司))**: The host cloud.tencent.com closed the connection on every attempt (curl exit 56, no HTTP status) from this session, so no page was read.
- **Vertex AI (Gemini Enterprise Agent Platform) (Google LLC)**: Express mode has a 90-day free tier with a subset of functions; its rate limits and whether billing details are needed weren't readable in the page text.

- not this job: Amazon Bedrock (Amazon Web Services, Inc.): Every route needs an AWS account with IAM or a Bedrock API key; there is no card-free API tier.
- not this job: Azure OpenAI in Foundry Models (Microsoft Corporation): An Azure subscription with a payment method, a resource and a deployment come before the first call; Tier 0 is a quota tier, not a card-free API.
- not this job: Cerebras Inference (Cerebras Systems Inc.): No permanent free tier; the $5 trial needs a verified payment method and expires after 30 days.
- not this job: DeepInfra (Deep Infra Inc.): No free tier; the account must add a card or prepay (pricing notes), and the rate-limit page describes paid accounts only.
- not this job: DeepSeek API (Hangzhou DeepSeek Artificial Intelligence Co., Ltd.): Prepaid balance with no free tier; 402 Insufficient Balance ends the run.
- not this job: Eden AI (Eden AI SAS): Prepaid credits bought by card, PayPal or bank transfer with a 5.5 per cent fee (pricing notes); no card-free API tier in the pages read.
- not this job: Glimpse (Elorian AI, Inc.): Paid per token on OpenRouter; free trial access is on Elorian's own platform through sales.
- not this job: Hugging Face Inference Providers (Hugging Face, Inc.): Free users have had no included credits since 7 October 2026 and must buy credits; PRO and Team seats get $2 a month.
- not this job: Meta Model API (Meta Platforms, Inc.): No free allowance; billing is in arrears to a card (pricing notes).
- not this job: Mistral AI API (Mistral AI (RCS Paris 952 418 325)): The usage-limits page says Free mode has no API access and that creating API keys and using the API require billing activation and pay as you go.
- not this job: Muse Spark Open (Meta Platforms, Inc.): Weights are free but need a four-GPU Blackwell node or a paid host; no free hosted endpoint is documented.
- not this job: Prism Inference (Prism Technologies Inc): Prepaid only with no free tier; agent sign-up is closed and the workspace owner must fund the account in a browser and hand over the key.
- not this job: RouterPlus (EVO HQ Inc): No free credit on sign-up and every key runs at 20 requests a minute until the organisation buys credit, so there is no card-free path.
- not this job: Together AI Inference (Together Computer, Inc.): A $15 credit purchase is required before platform access per the pricing notes, and 402 is returned at the spend cut-off; no free tier.
- not this job: Vercel AI Gateway (Vercel Inc.): The FAQ's 403 customer_verification_required says a team must add a valid payment method before using free credits, so there is no card-free path.
- not this job: xAI API (Grok) (SpaceXSI (xAI)): Tier 0 ($0 spend) sets limits but is not free credit; the only free trial in the docs is the consumer Grok app's Premium trial.

6 listings in the category were added after this reading and aren't marked yet.

## What stops agents finishing

- agent loops multiply calls per user turn so daily request caps go in minutes
- paid tier still returns errors citing the free-tier quota metric
- free models all return 429 under a burst

## Around it

After: inference.tool-loop.

## Where this comes from

Weight 4 of 100 in Model APIs & inference: 4 of 28 accounts, all failures of free or trial tiers under agent-style call volume

As data: https://www.anchorterminal.com/api/v1/use-cases/inference.free-start.json, or the MCP tool get_use_case with id inference.free-start. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
