Free AI Credits That Reset in 2026: 16 Allowances That Refill Daily, Weekly or Monthly
Signup credits run out. These AI APIs, GPU clouds and apps refill your free allowance on a schedule, so you can keep building for $0. Ranked by how much the recurring allowance actually gets you.
A $5 signup credit gets you through a weekend. A free allowance that refills every day or every month lets you run a side project, an internal tool or a test suite for as long as the provider keeps the offer going.
This post covers only that second kind: free allowances that reset on a schedule. One-time credits and time-limited trials are left out. For every free tier, recurring or not, see our broader guide to AI API free tiers.
Each entry lists the verified allowance, when it resets, and what you can realistically do with it. Entries are ranked within each group by how much useful work the allowance covers.
Last verified October 2026 against each provider’s official pricing or docs page. Free tiers change often; check the linked source before you build on one.
The comparison
| Provider | Free allowance | Resets | Best for |
|---|---|---|---|
| Google Gemini API | Free tier on most Flash models, Gemini 2.5 Pro and Gemma 4; limits shown in AI Studio | Daily (midnight PT) | General LLM prototyping |
| OpenAI (data sharing) | Up to 1M + 10M tokens/day for eligible orgs | Daily (00:00 UTC) | GPT models, if you can share traffic |
| Groq | 1,000 requests and 200K tokens/day per model (gpt-oss-120b) | Daily | Fast chat, Whisper transcription |
| Cloudflare Workers AI | 10,000 Neurons/day | Daily (00:00 UTC) | Small LLMs, embeddings, Flux Schnell |
| OpenRouter | 50 requests/day on :free models (1,000 after $10 lifetime top-up) | Daily | Trying free open models through one API |
| Cohere | 1,000 API calls/month (trial key) | Monthly | Embeddings and rerank prototypes |
| ElevenLabs | 10,000 credits/month | Monthly | Low-volume text-to-speech |
| Z.ai (Zhipu) | GLM-4.7-Flash, GLM-4.5-Flash, GLM-4.6V-Flash priced at $0 | Ongoing | Free GLM models |
| Token Harbor | Free allowance on promotional models | Rolling 7 days | Free DeepSeek/MiMo Flash calls |
| Modal | $30/month compute (Starter) | Monthly | Serverless GPU jobs and custom models |
| Hugging Face ZeroGPU | 5 GPU-minutes/day (free account) | 24h after first use | Gradio demos on shared GPUs |
| Google Flow | 50 credits/day | Daily | Short AI video clips |
| Jules | 15 tasks/day, 3 concurrent | Rolling 24h | Async coding agent tasks |
| GitHub Copilot Free | 2,000 completions + 50 chat requests/month | Monthly | Light IDE assistance |
| Lovable | 5 build credits/day, 30/month cap | Daily | Small app prototypes |
| Suno | 50 credits/day | Daily | Personal music generation |
API inference providers
1. Google Gemini API
Google’s free tier is still the broadest recurring LLM allowance from a frontier lab. The pricing page lists “Free of charge” input and output for current Flash and Flash-Lite models (including Gemini 3.8 Flash), Gemini 2.5 Pro, several TTS and Live models, and Gemma 4. Daily request quotas reset at midnight Pacific time.
The catch: Google no longer publishes per-model free limits in the docs. Your exact RPM, TPM and requests-per-day are shown only in AI Studio for your project. Free tier content is also used to improve Google’s products, so keep sensitive data out of it.
What you can realistically do: build and demo a full LLM app on a current Flash model without a billing account. Check your project’s limits in AI Studio before you plan any daily batch job around it.
2. OpenAI complimentary tokens for shared traffic
This one is easy to miss. If your organization opts in to share API inputs and outputs with OpenAI and is shown as eligible, you get daily complimentary tokens: up to 1M tokens/day on the larger model group and 10M/day on the mini/nano group for Launch and Grow organizations, or 250K and 2.5M for Build organizations. The counter resets at 00:00 UTC.
Conditions matter here. Not every organization is eligible, you need a positive account balance, fine-tuned models and tool use are excluded, and a request that would push you over the daily cap is billed in full. OpenAI says it will give 30 days’ notice before ending the program.
What you can realistically do: if you qualify and your data isn’t sensitive, run millions of mini-model tokens a day for evals, classification or synthetic data at no cost. That is the largest recurring allowance in this list.
3. Groq
Groq’s free plan gives per-model daily limits. For openai/gpt-oss-120b, gpt-oss-20b and qwen/qwen3.8-27b, that’s 30 requests/minute, 1,000 requests/day, 8K tokens/minute and 200K tokens/day each. Whisper Large v3 and v3 Turbo get 2,000 requests/day and 28,800 audio seconds/day (8 hours).
What you can realistically do: the 200K daily token cap binds long before the request cap. At roughly 2K tokens per call, budget about 100 calls per model per day. Because limits are per model, you can spread work across models. The 8 hours/day of Whisper transcription is the standout.
4. Cloudflare Workers AI
Workers AI includes 10,000 Neurons per day on both the Workers Free and Paid plans, resetting at 00:00 UTC. Neurons are Cloudflare’s unit of compute, priced per model.
Using Cloudflare’s published rates (our arithmetic): about 2.4M input tokens a day on Llama 3.1 8B fp8-fast, around 300 requests a day on gpt-oss-20b at 1K input plus 500 output tokens each, or roughly 170 Flux 1 Schnell images at 1024x1024 and 4 steps. On the free plan, requests fail once you hit the cap rather than billing you.
What you can realistically do: embeddings, classification and small-model chat inside a Worker, plus a modest daily image budget. Good for edge features that need a hard $0 ceiling.
5. OpenRouter
Models with a :free suffix on OpenRouter are capped at 20 requests/minute and 50 requests/day. If you have bought at least $10 of credits in total, the daily cap rises to 1,000 requests, and stays there.
The free model lineup rotates, so check what is currently free before depending on a specific model.
What you can realistically do: 50 requests a day is enough for testing and personal use. The $10 one-time purchase that unlocks 1,000 free requests a day is the best-value move in this whole post. See our free LLM APIs ranking for providers serving free models.
6. Cohere
Cohere trial keys are capped at 1,000 API calls a month, with 20 requests/minute on Chat, 10/minute on Rerank and 2,000 inputs/minute on Embed. Cohere positions trial keys for testing and proofs of concept, not production.
What you can realistically do: prototype a retrieval pipeline with Embed and Rerank. Since one Embed call can carry many inputs, 1,000 calls covers more text than the number suggests.
7. ElevenLabs
The ElevenLabs Free plan includes 10,000 credits a month with no rollover. On the API pricing page that equals 10,000 characters, about 10 minutes, on TTS v4. Flash/Turbo TTS costs half as much per character, so the same credits go further. Commercial licensing starts on the paid Starter plan.
What you can realistically do: voice a prototype or a few short demo clips each month. Not enough for regular content.
8. Z.ai (Zhipu): free-priced GLM models
Rather than a quota, Z.ai prices three models at $0 on its pricing page: GLM-4.7-Flash, GLM-4.5-Flash and the vision model GLM-4.6V-Flash. No rate limits are published for them.
What you can realistically do: use a capable free GLM model for development and light workloads. Since limits aren’t documented, test throughput before relying on it.
9. Token Harbor
Token Harbor offers a free allowance on promotional models, currently DeepSeek V4.1 Flash, MiMo V2.6 Flash and TH-Rudder, that resets on a rolling 7-day period. You enable the free models in your dashboard, and no card is required. The size of the allowance isn’t published.
What you can realistically do: free weekly access to fast DeepSeek and MiMo models through an OpenAI-compatible API, at a volume you’ll need to discover yourself.
Serverless and GPU clouds with recurring credits
Very few GPU clouds refill credits for free accounts. Most offer one-time signup bonuses (see our GPU cloud providers). Two that do:
10. Modal
Modal’s Starter plan has no monthly fee and includes $30 of free compute every month, with up to 10 concurrent GPUs. At Modal’s per-second rates, $30 is roughly 7.5 hours on an H100, 12 hours on an A100 80GB, 37 hours on an L4 or 50 hours on a T4 (our arithmetic).
What you can realistically do: host your own open-weight model, a Whisper or Flux endpoint, or nightly batch jobs that scale to zero. This is the most flexible recurring allowance in this post, because you choose the model. For a broader look at the category, see our serverless inference guide.
11. Hugging Face ZeroGPU
ZeroGPU Spaces run on shared NVIDIA RTX Pro 6000 Blackwell GPUs. Free accounts get 5 minutes of GPU time a day (2 minutes unauthenticated), reset 24 hours after first use, and accounts in good standing can host up to 2 ZeroGPU Spaces. Note that Hugging Face’s monthly Inference Providers credits go only to PRO and team users, not free accounts.
What you can realistically do: publish a Gradio demo of an image or LLM model that visitors run on their own quota. Five minutes a day is a few dozen short generations of your own.
Consumer platforms with daily credits
These are apps rather than APIs, but their allowances also refill.
- Google Flow: non-subscribers get 50 credits a day, refreshed the day after first use. A Veo 3.1 Lite clip costs 10 credits and Veo 3.1 Fast costs 20, so that’s 2 to 5 short clips a day. Video may be unavailable to free users at peak hours.
- Jules: Google’s async coding agent allows 15 tasks a day on a rolling 24-hour window, 3 at a time, on the base plan.
- GitHub Copilot Free: 2,000 inline completions and 50 chat requests a month. Enough for occasional use, not a full-time workflow.
- Lovable: 5 build credits a day, capped at 30 a month, plus 20 Cloud credits a month.
- Suno: 50 credits a day, with no commercial rights on the free plan.
For image generators with daily credits, see our free AI image platforms ranking.
Recurring offers we checked and left out
- Cerebras: our earlier listing had a standing free tier. Cerebras now offers a $5 credit that expires 30 days after activation, so it’s a trial.
- SiliconFlow: the pricing page now lists $1 of signup credit and no $0 models.
- Mistral: a free API tier has existed, but we couldn’t find current limits on an official Mistral page.
- Voyage AI: offers 200M free tokens per account on current embedding models, but the pricing page doesn’t say the tokens refill.
- GitHub Models: fully retired on July 30, 2026.
How to stack them
Use Gemini or Groq as your main free LLM and OpenRouter :free models as a fallback, with Cloudflare Workers AI for embeddings and small image jobs. Put anything custom on Modal’s $30 a month. Each allowance is independent, so a router with per-provider counters can keep a low-traffic app at $0 well past the point where signup credits would have run out.
When you need guaranteed throughput, paid rates are usually small at this scale. Compare them on Inference Hub’s provider pages.