Updated 21 November 2026. We checked every offer in the April version of this guide against the provider’s own page today. Several did not survive. Offers that we could not confirm have been removed or marked as ended, and every figure left has a source and a date. Free API offers change faster than any other kind of AI pricing, so treat this as a snapshot.

Today, three kinds of free API access are still real for developers: Google AI Studio has a free tier for some Gemini models, NVIDIA’s build.nvidia.com gives Developer Program members free API access to hosted models for prototyping, and Cerebras gives a one-time $5 promotional credit that expires after 30 days. Together AI, by its own billing page, no longer offers free trials and requires a first purchase of at least $15. Fireworks lists no free credits for new accounts. If you read an older article promising a $1 Together credit or a fixed monthly NVIDIA request count, those are no longer on the providers’ pages.

This guide is for developers who want to call a model from code without paying. If you want a free chat app, read our free plans guide. If you are deciding which paid plan to buy for plugin work, read our paid-plan comparison.

What changed since the April version of this guide?

OfferWhat we said in AprilStatus on 21 November 2026
Together AI signup credit$1 free credit, no cardNot confirmed. Together’s billing page says it “doesn’t currently offer free trials” and requires a first purchase of at least $15. Removed.
NVIDIA NIM free tier1,000 requests a monthThe monthly count is not on any NVIDIA page we could read. NVIDIA says Developer Program members get free API access for prototyping, with limits that vary by model. Count removed.
Google AI Studio free tier1,500 requests a day on Gemini 2.0 Flash, 50 a day on 2.5 Pro ExperimentalStill free for some models, but the model lineup and limits have changed. Those models are gone or deprecated. Numbers removed.
CerebrasFree tier, 60 requests an hour, no cardChanged. Cerebras now lists a one-time $5 promotional credit that needs a payment method and expires after 30 days. Rewritten.
Fireworks AI$1 trial credit, no cardNot confirmed. The pricing page mentions no free credits. One model is free until 30 November 2026. Rewritten.
GitHub Student Developer Pack AI creditsCredits from several AI providersNot verified on the provider page today. Removed.

Is Google AI Studio free in 2026?

Yes, for some models. Google’s AI Studio pricing page marks a free tier on Gemini 4 Hydrogen, Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.1 Flash-Lite, among others. It marks Gemini 4 Carbon, Argon and Neon as not available on the free tier, and it marks Gemini 2.5 Pro and 2.5 Flash as deprecated, with traffic rerouted to newer models on 1 December 2026. Paid prices for the models with a free tier are, per million input and output tokens: Gemini 4 Hydrogen $1.50 and $7.50, Gemini 3.8 Flash $0.75 and $3.75 through 31 December (then $1.50 and $7.50), and Gemini 3.5 Flash-Lite $0.30 and $2.50.

The rules of the free tier, from Google’s rate-limit page:

  • You need an active project or free trial, and some projects need verification.
  • Limits are measured in requests per minute, input tokens per minute and requests per day. Exceeding any one triggers an error even if you are under the others.
  • Limits are “applied per project, not per API key”.
  • Daily request quotas “reset at midnight Pacific time”.
  • The per-model free numbers are shown in your own AI Studio dashboard, not on the docs page, so we are not printing any. Look at your project’s active limits before building.

The data point that matters most to agencies: Google’s pricing page says free-tier content is used to improve its products, while paid usage generally is not. Do not send client code or customer data through a free key.

A first call, checked 21 November 2026

Google’s get-started page shows this REST call for the free-tier model Gemini 4 Hydrogen. Set the key as an environment variable, never in a plugin file or a repository.

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Api-Revision: 2026-05-20" \
  -d '{"model": "gemini-4-hydrogen", "input": "Write a PHPDoc block for a function that returns a sanitized slug."}'

The call format changed from the older generateContent pattern we showed in April. If you have scripts using the old format, check Google’s page before assuming they still work.

What are the NVIDIA NIM free tier limits?

NVIDIA does not publish a fixed number on the pages we could read. The build.nvidia.com home page says its models offer “Free inference with leading models” and lists four on the day we read it: nemotron-3.5-super-vl-120b-a12b, nemotron-3.5-lightning-30b-a3b, kimi-k3.1-pro and nemotron-3-ultra-550b-a55b. NVIDIA’s NIM FAQ on its developer forum, posted September 2024, says Developer Program members “have free access to NIM API endpoints for prototyping”, that the API Catalog is a trial with call limits “designed to serve evaluation and prototyping needs only”, that limits vary by model and may change with concurrent users, and that using NIM in production requires an NVIDIA AI Enterprise license. To see your own limit for a model, the FAQ says to check your account at the top right of build.nvidia.com.

The rate figure developers quote is about 40 requests per minute, but that comes from forum posts, not from NVIDIA documentation, and many threads ask NVIDIA to raise it to 200. See one such request. Treat 40 as an anecdote and read your own account page.

Best free NVIDIA NIM models for coding

We would not rank them without testing them on your code, and we have not benchmarked these four. What can be said from the page: the catalog changes often, so choose from what build.nvidia.com lists on the day, prefer the models described as coding or reasoning models, and run the same three or four prompts on each. A model with the right size for your latency budget and a license you can live with matters more than a leaderboard position. Remember the production restriction above: this is for prototyping.

Does Together AI give free credits?

On 21 November 2026, no. Together’s billing documentation says it “doesn’t currently offer free trials” and that your organization’s first purchase must be at least $15, with free credits on your balance not counting toward that minimum. The pricing page lists one model, Ternary Bonsai 27B, as free for input and output, which is a model price, not a signup credit. Other serverless prices on the page, per million input and output tokens: gpt-oss-120B $0.15 and $0.60, Qwen3.5 9B $0.17 and $0.25, DeepSeek V4.1 Flash $0.30 and $1.20, GLM-5.3 $1.40 and $4.40 and Kimi K3 $3.00 and $15.00. The page’s metadata says “Start for free”, but the billing page is more specific, so we went with that.

Is there free Cerebras credit?

Cerebras lists a one-time $5 promotional credit on its pricing page. To receive it you add a valid payment method, and the page says you will not be charged or automatically moved to paid usage. The credit expires 30 days after activation, and API and Playground access pause when it expires or runs out unless you buy pay-as-you-go credits. Its rate-limit page lists Free Trial limits: gpt-oss-120b at 5 requests per minute with a 1M-token daily cap, and qwen-3.8-27b at 300 requests per minute with no hourly or daily token caps. It says there is no permanent free tier. This is a 30-day evaluation, not a free plan.

Does Fireworks AI have free credits?

Its pricing page does not mention free credits or a free tier, and says serverless usage draws down prepaid credits. The only free offer we saw was a single model, Muse Glimmer 30B SFT, listed as free through 30 November 2026. Fireworks’ rate-limit page lists an account-wide limit of 10 requests per minute for accounts with no payment method or no credits. If you are reading this after 30 November, assume the free model offer has ended.

Is there free Claude, OpenAI or Grok API credit?

Not as a standing offer that we could confirm. xAI’s models page lists current prices, for example grok-4.3 at $1.25 and $2.50 per million tokens for prompts under 200k, and mentions no free credits or data-sharing programme. Anthropic’s pricing page lists Max with $100 a month in API credits, which is a paid-plan benefit, not a free offer. OpenAI’s pricing page mentions API credits only inside the Pro tiers. Check each provider’s console for promotional credits on new accounts, because those are time-limited and can be regional.

Is there free Claude access for students?

Not for individual students as a standard offer, as far as Anthropic’s pages show. Anthropic’s education page describes a university-wide plan for institutions, including students, faculty and staff, sold through an Education team, with no student pricing or terms listed. The same page offers free access to start for scientists at academic and nonprofit research institutions. Students without a campus agreement can use the free Claude plan, which we cover in our free plans guide. Ask your school whether it has an agreement before assuming it does not.

Which free AI API is best for coding?

It depends on what you are doing, and none of them replaces a paid account for production traffic.

  • Scripts, prototypes and docs generation: Google AI Studio, on a free-tier model, as long as the data is not sensitive.
  • Trying large open models for a prototype: NVIDIA’s hosted catalog, within the Developer Program’s prototyping terms.
  • A quick 30-day speed test: Cerebras, using the $5 credit, if you are happy to add a payment method.
  • Anything that touches client data or production traffic: a paid API with data terms you have read. Our LLM cost guardrails post covers the budget controls to put in place before you start paying.

How do you use free API tiers for WordPress work?

The jobs that fit a free API are small, repeatable and not sensitive: generating PHPDoc comments for a plugin you own, drafting readme sections, classifying support tickets from public text, summarising a changelog, or sketching a REST schema. The jobs that do not fit are anything that sends customer data, anything that must run reliably in production, and anything that needs a model the free tier excludes. Two habits keep it cheap and safe: keep the key in an environment variable and write a small wrapper function that takes a provider name, so switching providers when an offer ends is a one-line change. Our Aider and Gemini Flash-Lite post shows one way to wire a cheap Gemini model into a coding workflow, though it predates the current Gemini lineup, so substitute a model from the free-tier list above.

How do you keep a provider wrapper so you can switch when an offer ends?

Free offers end, so write your code so that the provider is a setting, not a hard-coded URL. This is a skeleton in the shape we use for scripts that live next to a plugin repository, checked on 21 November 2026 only for the Gemini header names, which we copied from Google’s get-started page. Everything else is a placeholder to fill from the provider’s own docs.

/**
 * Skeleton only. Endpoints, headers and response shapes differ per provider,
 * so fill in each provider from its own docs before using it.
 */
function acme_ai_request( string $provider, array $payload ): array|WP_Error {
	$providers = array(
		'gemini' => array(
			'url' => 'https://generativelanguage.googleapis.com/v1beta/interactions',
			'key' => getenv( 'GEMINI_API_KEY' ),
		),
	);

	if ( empty( $providers[ $provider ]['key'] ) ) {
		return new WP_Error( 'acme_ai_no_key', 'No API key set for ' . $provider );
	}

	$response = wp_remote_post(
		$providers[ $provider ]['url'],
		array(
			'timeout' => 30,
			'headers' => array(
				'x-goog-api-key' => $providers[ $provider ]['key'],
				'Content-Type'   => 'application/json',
				'Api-Revision'   => '2026-05-20',
			),
			'body'    => wp_json_encode( $payload ),
		)
	);

	if ( is_wp_error( $response ) ) {
		return $response;
	}

	return json_decode( wp_remote_retrieve_body( $response ), true ) ?: array();
}

Keep the key in the environment of the machine that runs the script, not in the database and not in the repository. If you must store it in WordPress, put it in wp-config.php as a constant outside version control.

How do you evaluate any free API offer before you build on it?

  1. Find the offer on the provider’s own billing or pricing page. A blog post, a video or a forum answer is a lead, not a source. If the provider’s page does not say it, treat the offer as unconfirmed.
  2. Write down the date and the exact wording. “One-time credit”, “free through 30 November” and “free tier” are three different promises.
  3. Check the expiry and the payment method. Cerebras’s credit needs a card on file and ends after 30 days. Some offers pause your access instead of charging you, and some do not.
  4. Check the data terms. Google’s pricing page says free-tier content is used to improve products. If you cannot send the data you plan to use, the offer is not for that job.
  5. Check the production terms. NVIDIA’s FAQ limits Developer Program access to prototyping, research, development and testing.
  6. Find the exit price. Look at what the same model costs once you pay. If it costs more than you can afford for the volume you plan, free is just a delay.

What does a small job cost once you leave the free tier?

Here is the arithmetic, with an assumed job so you can substitute your own. Suppose a script sends about 2,000 input tokens and gets back about 500 output tokens per plugin file, and you run it on 300 files. That is 600,000 input tokens and 150,000 output tokens.

  • On Gemini 3.5 Flash-Lite at $0.30 and $2.50 per million tokens, that is about $0.18 for input and $0.375 for output, so roughly 55 cents.
  • On Gemini 4 Hydrogen at $1.50 and $7.50, it is $0.90 plus $1.125, about $2.03.
  • On Together’s gpt-oss-120B at $0.15 and $0.60, it is $0.09 plus $0.09, about 18 cents, though you must make the $15 minimum first purchase.

The point of the arithmetic: for small developer jobs, paid prices on cheap models are low enough that the free tier mostly saves you setup friction, not money. It is worth using free tiers to learn which model suits the job, and worth moving to a cheap paid model when reliability matters. The $15 Together minimum is the bigger decision than the per-token price.

What should an agency never send to a free API?

Because the tier you use can train on, or at least retain, what you send, set a short list of things that never go through a free key:

  • Customer personal data from a client site, including exports, support tickets with names and order details.
  • Credentials of any kind: API keys, database passwords, licence keys, application passwords.
  • Unreleased client code or content covered by a contract that restricts third-party processing.
  • Your own unreleased product code, if you would not want it appearing in a provider’s training data.

What is left is still plenty: public documentation, your open-source plugin code, synthetic test data and text you have already published. Put the rule in your team notes next to the key-handling rule, so a new developer finds both on day one. If a task needs more than that, pay for an API whose terms you have read, and set a spending limit before the first call.

One more habit is worth adopting. Keep a small file in the repository, next to the script, that records which offer you used, the date you checked it, the page you checked it on and the limit you saw. When the script starts returning errors in three months, that file tells you in a minute whether the offer changed or your code did, and it gives the next person on the team a place to start instead of a guess.

Common mistakes people report

Checked 21 November 2026, paraphrased from provider forums and pages.

  • Getting a 429 right after creating a key. Developers on Google’s forum report a new free-tier account returning 429 immediately and a new free-tier project hitting 429. Check that the model you picked has a free tier at all and that the project is active.
  • 429s on NVIDIA despite low usage. Users describe 429 errors on NVIDIA NIM despite low usage. Limits vary by model and with concurrency, so retry with backoff and keep one request in flight at a time.
  • Shipping a prototype on a prototyping tier. NVIDIA’s FAQ says production use needs an AI Enterprise license.
  • Missing the credit expiry. Cerebras credit expires 30 days after activation, and access pauses when it ends.
  • Assuming a free key keeps your data private. Google’s pricing page says free-tier content is used to improve its products.

Frequently asked questions

Does Together AI give free credits?

Not according to its billing page today. It says it does not offer free trials and requires a first purchase of at least $15.

Is NVIDIA NIM free?

Developer Program members get free API access to hosted models for prototyping, with limits that vary by model. Production use needs an NVIDIA AI Enterprise license.

Is the Gemini API free?

For some models, yes. Limits apply per project, daily quotas reset at midnight Pacific time, and free-tier content may be used to improve Google’s products. The newest top models are not on the free tier.

What happened to Cerebras’s free tier?

Cerebras now offers a one-time $5 promotional credit with a payment method on file, expiring after 30 days, and says there is no permanent free tier.

Can I use a free API for client work?

Not for client data on a tier that uses content to improve products, and not for production on a prototyping tier. Use a paid API with terms you have read.

How do I know an offer is still live?

Open the provider’s own billing or pricing page and look for the date and the wording. If you cannot find the offer there, assume it has ended.