The LLM API free tier pack: what's in it and how to use it
Model and embedding APIs you can call for free today: free tiers, daily quotas and small signup credits. The pack now has 28 offers; Scaleway Generative APIs joined it today. Who it's for: Anyone prototyping with LLMs. No application needed, though some need a payment method on file.
The offers
- Google (Gemini API) (Free of charge input/output on Free Tier models (rate-limited))
- Groq (Free Plan with per-model rate limits (RPM/RPD/TPM/TPD))
- Cohere (Free Trial API key (rate-limited; 1,000 API calls/month cap))
- OpenRouter (Free :free model variants — 50 requests/day (1,000/day after buying 10 credits), 20 RPM)
- Cloudflare (10,000 Neurons per day free on Workers AI)
- OpenAI (Free daily token quota (e.g. 1M tokens/day on eligible models) when sharing traffic with OpenAI)
- Cerebras (One-time $5 promotional credit (expires 30 days after activation))
- AI21 Labs ($10 credits for 7 days, no credit card)
- Fireworks AI ($1 starter credit)
- Vercel ($5/month AI Gateway credit included on every team)
- Voyage AI (MongoDB) (First 200 million tokens free per account (voyage-4 family embeddings))
- Jina AI (10M free tokens with every new API key)
- SiliconFlow ($1 in free credits to get started)
- Portkey (Developer plan free forever: 10k recorded logs per month)
- Nebius (Free AI Builder Program membership: $25 Token Factory credit, plus partner credits ($25 Tavily, $100 LangSmith, $100 Toloka))
- Hetzner (Inference API free of charge while it is in experimental status)
- OpenCode (Rotating free models on OpenCode Zen (limited-time free coding models))
- Fireworks AI (Muse Glimmer 30B SFT free through November 30 (UI + Serverless Training API))
- Inception (100 million free tokens with every new Inception API key (Mercury models))
- NVIDIA (Free prototyping on build.nvidia.com model endpoints, up to 40 requests per minute for most models, no per-token billing)
- Maritaca AI (S5-free model: tokens not billed for a limited time (text only))
- SambaNova (SambaCloud Free Tier: rate-limited API access with daily token limits, no payment method needed)
- Nscale ($5 free credit for serverless inference)
- Alibaba Cloud (Free token quota for many models (Qwen and others) when you first activate Model Studio, valid 90 days)
- LG AI Research (EXAONE) (K-EXAONE 2.5 API free of charge for now (OpenAI-compatible, rate-limited))
- Z.ai (Zhipu) (GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash: free input and output tokens on the Z.ai API)
- Together AI (Ternary Bonsai 27B free on Together serverless ($0 input and output tokens))
- Scaleway (Free tier: first 1,000,000 tokens plus 60 minutes of audio transcription on Generative APIs)
A sensible order to claim them
- Start with the ones that only need an account. Gemini API (an AI Studio key on a free tier project), Groq, OpenRouter's
:freemodels, Cohere trial keys, Together's free Bonsai model, Z.ai's Flash models and the EXAONE API all work with a plain sign-up. Rate limits are the catch, not money, so they suit prototypes and evals. - Add the one-time token and credit grants next. Voyage (first 200 million tokens per account), Jina (10M tokens per new key), Inception (100 million tokens per new key), AI21 (credits for 7 days, no card), Scaleway (first 1,000,000 tokens) and the small Fireworks, SiliconFlow and Nscale starter credits. Some expire, so claim them when you are ready to build, not before.
- Use what you already pay for. If you have a Vercel team, the AI Gateway credit is already there each month. Cloudflare Workers AI gives a daily Neuron allowance on Workers Free or Paid. Hetzner customers can call its Inference API free while it is experimental.
- Check the conditions on the rest. Cerebras' promo credit needs a payment method on file. SambaNova's Free Tier applies only while no payment method is linked. OpenAI's free daily tokens need data sharing enabled on an eligible org. Alibaba's free quota is for new Model Studio users in the Singapore region. Fireworks' Muse Glimmer and Maritaca's S5-free are time-limited, and OpenCode Zen's free models rotate.
Every card above links to the vendor's own page and shows when we last checked it. Limits change often, so read the source quote before you plan around a number.
Originally published on Perq, where every listing carries a source link and a last-verified date. Perq is run by Avatar 8 LLC.
About the author
- Real email inboxes for AI agents, via API and MCP. By Avatar 8 LLC.
Comments
Join the discussionNo comments yet. Be the first to comment.