macrostack

AI API price index · free use · read 17 Nov 2026

Free LLM APIs, and the catch with each

12 of the 32 providers in the AI API price index let you use a language model free through the API. 6 have no end date stated — Cloudflare Workers AI, Google Gemini, Groq, Novita AI, Z.ai and Cohere; the rest run out or need a purchase first. Every limit below is quoted from the provider’s own page, read 17 Nov 2026.

What free means here

A free tier runs on rate limits with no end date stated. Free models are printed at a price of zero. A trial key is free but not for production. A credit is money on account: it runs out, and some expire. A credit is never free forever, and this page never calls it that.

Ordered by what the free use is, then by name. Nobody pays to be here or to move.

Free LLM APIs at a glance

12 providers, read 17 Nov 2026. The catch is in each provider's own words below.

ProviderWhat is freeHow long
Cloudflare Workers AIFree tier10,000 Neurons a day free on both Workers plans, reset at midnight UTC; past that, Workers Paid charges $0.011 per 1,000 Neurons. gpt-oss-120b uses 68,182 Neurons per million output tokens, so the daily allowance covers about 146,000 of its output tokens (our arithmetic from Cloudflare's figures). Nine models, the Kimi K2 and GLM 5.3 models among them, need a paid billing method.No end date stated
Google GeminiFree tierA Free tier with free input and output tokens on most Gemini models, under rate limits Google shows in each account's AI Studio. Gemini 4 Carbon, Gemini 4 Argon and Gemini 3.1 Pro Preview are not on it. What you send on the free tier is used to improve Google's products.No end date stated
GroqFree tierA Free plan defined by rate limits alone: GPT OSS 120B, GPT OSS 20B and Qwen3.8 27B at 30 requests a minute and 1,000 a day, with no dollar amount attached.No end date stated
Novita AIFree modelsTwo language models are printed as Free for input and output. No free credit or free trial is stated.No end date stated
Z.aiFree modelsThree GLM models are printed as Free for input, cached input and output. The pricing page states no rate limit for them and no free credit.No end date stated
Together AIFree modelsTwo models are priced at zero, but nothing is free to start: there is no free trial, and the platform needs a first credit purchase of at least $15.After a first purchase
CohereTrial keyEvery account gets a free Trial API key: 20 chat requests a minute and 1,000 API calls a month, and not for production or commercial use.No end date stated
Alibaba Cloud Model Studio (Qwen)One-time creditA free quota of 1 million tokens on each of several models, valid for 90 days, and only in the Singapore region.One-time: it runs out
AnthropicOne-time creditNew users get free credits to test the API; the amount is not stated, and the price table has no free tier.One-time: it runs out
CerebrasOne-time creditA one-time $5 credit once you add a payment method. It expires 30 days after activation, and access pauses when it runs out unless you buy credits. The tier it opens is for development, not production.One-time: it runs out
Fireworks AIOne-time creditThe pricing page offers no free tier; Fireworks' billing FAQ is titled "What happens when I finish my $1 credit?", so a $1 starting credit is implied rather than offered, and the account is suspended when it is gone unless a payment method is added.One-time: it runs out
ReplicateFree for a whileSelect models run free for a while before Replicate asks you to set up billing. Which models, and for how long, is not stated.For a while: no amount stated

Each free LLM API, in the provider's words

Cloudflare Workers AI Free tier · No end date stated

10,000 Neurons a day free on both Workers plans, reset at midnight UTC; past that, Workers Paid charges $0.011 per 1,000 Neurons. gpt-oss-120b uses 68,182 Neurons per million output tokens, so the daily allowance covers about 146,000 of its output tokens (our arithmetic from Cloudflare's figures). Nine models, the Kimi K2 and GLM 5.3 models among them, need a paid billing method.

Named as free: @cf/openai/gpt-oss-120b, @cf/openai/gpt-oss-20b.

  • “Our free allocation allows anyone to use a total of 10,000 Neurons per day at no charge.”
  • “Workers AI is included in both the Free and Paid Workers plans and is priced at $0.011 per 1,000 Neurons.”
  • “To use more than 10,000 Neurons per day, you need to sign up for the Workers Paid plan.”
  • “All limits reset daily at 00:00 UTC.”
  • “Some models require a paid billing method.”
  • @cf/openai/gpt-oss-120b: 31818 neurons per M input tokens · 68182 neurons per M output tokens

Read 17 Nov 2026 on developers.cloudflare.com/workers-ai/platform/pricing

Google Gemini Free tier · No end date stated

A Free tier with free input and output tokens on most Gemini models, under rate limits Google shows in each account's AI Studio. Gemini 4 Carbon, Gemini 4 Argon and Gemini 3.1 Pro Preview are not on it. What you send on the free tier is used to improve Google's products.

Named as free: Gemini 4 Hydrogen, Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite, Gemini 3 Flash Preview, Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-Lite, Gemini Embedding.

Not on the free tier: Gemini 4 Carbon, Gemini 4 Argon, Gemini 3.1 Pro Preview.

  • “For developers and small projects getting started with the Gemini API.”
  • “Limited access to certain models”
  • “Free input & output tokens”
  • “Content used to improve our products”
  • “Rate limits depend on a variety of factors (such as your usage tier) and can be viewed in Google AI Studio.”

Read 17 Nov 2026 on aistudio.google.com/docs/pricing · aistudio.google.com/docs/rate-limits

Groq Free tier · No end date stated

A Free plan defined by rate limits alone: GPT OSS 120B, GPT OSS 20B and Qwen3.8 27B at 30 requests a minute and 1,000 a day, with no dollar amount attached.

Named as free: openai/gpt-oss-120b, openai/gpt-oss-20b, qwen/qwen3.8-27b, whisper-large-v3, whisper-large-v3-turbo.

  • “Upgrade to Developer plan to access higher limits, Batch and Flex processing, and more.”
  • Free Plan Limits — openai/gpt-oss-120b: RPM 30 · RPD 1K · TPM 8K · TPD 200K
  • Free Plan Limits — openai/gpt-oss-20b: RPM 30 · RPD 1K · TPM 8K · TPD 200K
  • Free Plan Limits — qwen/qwen3.8-27b: RPM 30 · RPD 1K · TPM 8K · TPD 200K
  • Free Plan Limits — whisper-large-v3: RPM 20 · RPD 2K · ASH 7.2K · ASD 28.8K

Read 17 Nov 2026 on console.groq.com/docs/rate-limits

Novita AI Free models · No end date stated

Two language models are printed as Free for input and output. No free credit or free trial is stated.

Named as free: zai-org/glm-new, Apodex 1.1 Mini.

  • “Free to start, scales as you grow.”
  • zai-org/glm-new: input Free · output Free
  • Apodex 1.1 Mini: input Free · output Free

Read 17 Nov 2026 on novita.ai/pricing

Z.ai Free models · No end date stated

Three GLM models are printed as Free for input, cached input and output. The pricing page states no rate limit for them and no free credit.

Named as free: GLM-4.7-Flash, GLM-4.5-Flash, GLM-4.6V-Flash.

  • GLM-4.7-Flash: Free | Free | Free
  • GLM-4.5-Flash: Free | Free | Free
  • GLM-4.6V-Flash: Free | Free | Free

Read 17 Nov 2026 on docs.z.ai/guides/overview/pricing

Together AI Free models · No end date stated

Two models are priced at zero, but nothing is free to start: there is no free trial, and the platform needs a first credit purchase of at least $15.

Pay first: “To use the platform, you must purchase at least $15 of credits.”

Named as free: Ternary Bonsai 27B, Tev1 4B Experimental (output only).

  • “Together AI doesn't currently offer free trials.”
  • “To use the platform, you must purchase at least $15 of credits.”
  • “The minimum is $15 for your organization's first purchase and $5 after that.”
  • Ternary Bonsai 27B: input Free · output Free
  • Tev1 4B Experimental: input $0.042 · output $0.00

Read 17 Nov 2026 on together.ai/pricing · docs.together.ai/docs/billing-credits

Cohere Trial key · No end date stated

Every account gets a free Trial API key: 20 chat requests a minute and 1,000 API calls a month, and not for production or commercial use.

  • “API calls made from a Trial API key are free. However, trial keys are rate limited and are not permitted to be used for production or commercial purposes.”
  • “Every account begins as a personal account and only has access to Trial API keys.”
  • “Trial keys (and prod keys on newer Chat model variants) are limited to 1,000 API calls a month.”
  • Trial rate limit, Chat (Command A+, Command A, Command R+, Command R7B and the rest): 20 req / min

Read 17 Nov 2026 on cohere.com/pricing · docs.cohere.com/docs/rate-limits

Alibaba Cloud Model Studio (Qwen) One-time credit · One-time: it runs out

A free quota of 1 million tokens on each of several models, valid for 90 days, and only in the Singapore region.

Named as free: qwen4-max, qwen3.8-max, qwen3.7-max, qwen3.7-plus, qwen3.8-flash, qwen3.7-flash, deepseek-v4.1-flash, glm-5.3.

  • “Valid for 90 days from the date of Model Studio activation, model release, or application approval, whichever is later”
  • “The following models offer a free quota only in Singapore. No free quota is available in other regions.”
  • Free quota: 1 million tokens per listed model

Read 17 Nov 2026 on alibabacloud.com/help/en/model-studio/model-pricing

Cerebras One-time credit · One-time: it runs out

A one-time $5 credit once you add a payment method. It expires 30 days after activation, and access pauses when it runs out unless you buy credits. The tier it opens is for development, not production.

  • “Add a valid payment method to receive a one-time $5 promotional credit.”
  • “The credit expires 30 days after activation.”
  • “You will not be charged or automatically enrolled in paid usage.”
  • “API and Playground access will pause when the credit expires or is exhausted unless you separately purchase PayGo credits.”
  • “For development, evaluation, and experimentation; not intended for production use.”

Read 17 Nov 2026 on cerebras.ai/pricing

Replicate Free for a while · For a while: no amount stated

Select models run free for a while before Replicate asks you to set up billing. Which models, and for how long, is not stated.

  • “You can run select models on Replicate for free, but after a bit you'll be asked to set up billing.”
  • “Some features are only available to customers with billing set up.”

Read 17 Nov 2026 on replicate.com/docs/topics/billing

Other AI APIs with free use

Embeddings, speech, voice and moderation models — free use of a model through the API, but not a language model.

Mistral AI Free models · No end date stated

Only the moderation model is free: Mistral Moderation 2 is printed as Free. There is no free API tier; the Free consumer plan includes Mistral Studio's playgrounds, not API credits.

Named as free: Mistral Moderation 2.

  • “On the Free plan, you also get access to Mistral Studio.”
  • Mistral Moderation 2: input Free · cached input Free · output Free

Read 17 Nov 2026 on docs.mistral.ai/inference/pricing · mistral.ai/pricing

OpenAI Free models · No end date stated

Only the moderation model is free: omni-moderation-latest is printed as Free. There is no free tier and no free credit for text models.

Named as free: omni-moderation-latest.

  • Moderation — omni-moderation-latest: Free

Read 17 Nov 2026 on developers.openai.com/api/docs/pricing

Cartesia Monthly free plan · Renews each month

A Free plan at $0 a month with 20,000 credits and $1 of prepaid agents a month — about 27 minutes of Sonic-3.6 text to speech. The commercial-use licence starts on Pro.

  • “20K credits / month”
  • “$1 prepaid agents / month”

Read 17 Nov 2026 on cartesia.ai/pricing

ElevenLabs Monthly free plan · Renews each month

A Free plan at $0 a month with 10,000 credits a month that do not roll over. A commercial licence first appears on the Starter plan.

  • “$0/mo”
  • “10k credits per month”
  • “Rollover does not apply to the Free plan.”

Read 17 Nov 2026 on elevenlabs.io/pricing · elevenlabs.io/pricing/api

AssemblyAI One-time credit · One-time: it runs out

Up to 185 hours of pre-recorded and 333 hours of streaming transcription free, with no card needed, then pay as you go.

  • “Start free, pay-as-you-go after that – no commitments required.”
  • “You can create an account and start transcribing immediately with no credit card required. The free tier includes up to 185 hours of pre-recorded transcription and up to 333 hours of streaming transcription.”

Read 17 Nov 2026 on assemblyai.com/pricing

Deepgram One-time credit · One-time: it runs out

$200 of free credit for speech, with no card needed. It does not expire, but it runs out: Deepgram puts it at roughly 43,000 minutes of Nova-3 transcription.

  • “Start for free with $200 credit.”
  • “No minimums, no expiration, and no credit card required.”
  • “Roughly 43,000 minutes, over 700 hours, of Nova-3 transcription.”
  • “It does not expire, so you can prototype without a clock running.”

Read 17 Nov 2026 on deepgram.com/pricing

Speechmatics One-time credit · One-time: it runs out

$100 of sign-up credit that lasts 12 months; on the free tier, API access pauses once it runs out. Text to speech has its first million characters free.

  • “Start free • $100 in credits”
  • “Sign-up bonus, credit-pack credits and annual subscription credits last 12 months.”
  • “If you’re on the free tier, your API access will be paused once your credits run out.”
  • “First 1 million characters free (about 20 hours of speech).”

Read 17 Nov 2026 on speechmatics.com/pricing

Voyage AI One-time credit · One-time: it runs out

Free embedding tokens for every account: the first 200 million on the current models, 50 million on several others. Older models get none, and free tokens do not apply to the Batch API.

  • “The first 200 million tokens for voyage-4-large, voyage-4, voyage-4-lite, voyage-context-4, and voyage-code-3, or the first 50 million tokens for voyage-multilingual-2, voyage-finance-2, voyage-law-2, and voyage-code-2, are free for every account.”

Read 17 Nov 2026 on docs.voyageai.com/docs/pricing

No free use stated

12 providers in the index state no free use of a model. Here is what each page does say.

ProviderWhat its page says
Amazon BedrockNo free allowance for model inference. "Free" on the page applies only to some Guardrails filters, and the AWS Free Tier is an account-wide offer, not a Bedrock price.Read 17 Nov 2026 on aws.amazon.com/bedrock/pricing
Azure AI Foundry ModelsNo free token allowance is printed on the Foundry Models pricing pages; Azure's free account credit is an account-wide offer.Read 17 Nov 2026 on azure.microsoft.com/en-us/pricing/details/microsoft-foundry
Azure OpenAINo free token allowance for Azure OpenAI models. The page's one free allowance is 1 GB of File Search storage; "Try Azure for free" is an Azure-wide account offer, not an Azure OpenAI price.Read 17 Nov 2026 on azure.microsoft.com/en-us/pricing/details/azure-openai
DeepInfraNo free tier or free credit: a card or a prepayment comes first.Read 17 Nov 2026 on deepinfra.com/pricing
DeepSeekNo free tier and no free credit are stated. Fees come out of a topped-up or granted balance; no granted amount is printed.Read 17 Nov 2026 on api-docs.deepseek.com/quick_start/pricing
Google Vertex AI (Agent Platform)No free token allowance is printed. Grounding with Google Search includes 5,000 queries a month at no charge, but that is a tool, not a model; Google Cloud's free trial credit is an account-wide offer.Read 17 Nov 2026 on cloud.google.com/model-garden/pricing
MiniMaxNo free tier or free credit: with pay-as-you-go billing, usage is deducted from a balance you add funds to.Read 17 Nov 2026 on platform.minimax.io/docs/pricing/overview
Moonshot AI (Kimi)No free tier: the API needs a $1 first recharge. A $5 voucher comes only after $5 of recharges, so it is a bonus on money spent, not free use.Read 17 Nov 2026 on platform.kimi.ai/docs/pricing/chat
Nebius Token FactoryNo free tier or free credit is stated; usage is charged to a billing account with a bank card.Read 17 Nov 2026 on tokenfactory.nebius.com/endpoints
PerplexityNo free tier or free credit is stated; credits are bought in the API console.Read 17 Nov 2026 on docs.perplexity.ai/docs/getting-started/pricing
SambaNova CloudNo free credit: the Free plan needs a payment method and a credit purchase before the first request.Read 17 Nov 2026 on cloud.sambanova.ai/plans/pricing
xAINo free tier, free credit or free model is stated on the API pricing page.Read 17 Nov 2026 on docs.x.ai/developers/pricing

Routers are not providers

OpenRouter passes other providers’ models and prices through, so it is not in the index and not in the list above. Its own free plan, read 17 Nov 2026: The Free plan has no fee and opens 25+ free models from 4 free providers. Models ending in :free allow 20 requests a minute, and 50 a day until you have bought 10 credits in all; from then on, 1,000 a day. The Free plan has no auto-routing, budgets, prompt caching or bring-your-own-key. Is OpenRouter free? The limits and the fee.

When free runs out

Questions

Is there a free LLM API?
Yes. On 17 Nov 2026, 12 of the 32 providers in Macrostack's AI API price index let you use a language model free through the API. 6 have no end date stated: Cloudflare Workers AI (free tier), Google Gemini (free tier), Groq (free tier), Novita AI (free models), Z.ai (free models) and Cohere (trial key). The other 6 are a one-time credit or quota, a free stretch with no amount stated, or free models behind a first purchase. Every limit is quoted from the provider's own page.
Is a free API credit free forever?
No. A credit runs out, and some expire. Alibaba Cloud Model Studio (Qwen): A free quota of 1 million tokens on each of several models, valid for 90 days, and only in the Singapore region. Anthropic: New users get free credits to test the API; the amount is not stated, and the price table has no free tier. Cerebras: A one-time $5 credit once you add a payment method. It expires 30 days after activation, and access pauses when it runs out unless you buy credits. The tier it opens is for development, not production. Fireworks AI: The pricing page offers no free tier; Fireworks' billing FAQ is titled "What happens when I finish my $1 credit?", so a $1 starting credit is implied rather than offered, and the account is suspended when it is gone unless a payment method is added. Replicate: Select models run free for a while before Replicate asks you to set up billing. Which models, and for how long, is not stated.
Can I use a free LLM API in production?
Check each provider's terms first. Google Gemini: "Content used to improve our products" Together AI: "To use the platform, you must purchase at least $15 of credits." Cohere: "trial keys are rate limited and are not permitted to be used for production or commercial purposes." Cerebras: "For development, evaluation, and experimentation; not intended for production use." Where a provider states its rate limits, they are quoted on this page.
What about OpenRouter's free models?
OpenRouter is a router: it passes other providers' models and prices through, so it is not a provider in this index. In its own words, read 17 Nov 2026: The Free plan has no fee and opens 25+ free models from 4 free providers. Models ending in :free allow 20 requests a minute, and 50 a day until you have bought 10 credits in all; from then on, 1,000 a day. The Free plan has no auto-routing, budgets, prompt caching or bring-your-own-key.
Does Macrostack get paid by any of these providers?
No. The list is ordered by what the free use is, then by name, and nobody pays to be on it or to move on it. There are no affiliate or sponsored links on this page.

Kinds used on this page: Free tier — an ongoing free tier, held back by rate limits — no end date stated; Free models — models printed at a price of zero; Trial key — a free key that is not for production or commercial use; One-time credit — a one-time credit or free quota — it runs out, and some expire; Free for a while — free for a while, with no amount stated; Monthly free plan — a free plan whose credits renew each month.

The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.