Compare · 8 providers · checked
Free AI API tiers
Rate limits, daily caps, card and training terms. Groq, OpenRouter, Cloudflare and Cohere publish their free limits; Google and Mistral only show them in your console.
Compared
Best free text model per provider · official docs only
30 RPM · 1K RPD · 8K TPM
Cap: 1K req · 200K tokens
Card: unknown · Trains on it: No
20 RPM · 50 RPD · unknown TPM
Cap: 50 req (1,000 after $10)
Card: No · Trains on it: Per provider
300 RPM · unknown RPD · unknown TPM
Cap: 10K Neurons
Card: unknown · Trains on it: No
20 RPM · unknown RPD · unknown TPM
Cap: 1,000 calls a month
Card: No · Trains on it: Yes
unknown RPM · unknown RPD · unknown TPM
Card: No · Trains on it: Yes
unknown RPM · unknown RPD · unknown TPM
Card: No · Trains on it: Unclear
NVIDIA NIM (build.nvidia.com)prototyping only
unknown RPM · unknown RPD · unknown TPM
Card: unknown · Trains on it: unknown
Cerebras (free trial)trial, not free
5 RPM · unknown RPD · 30K TPM
Cap: 1M tokens (30-day trial)
Card: Yes · Trains on it: unknown
Limits differ per model; see each provider below. Sort by any column.
Provider by provider
GroqCloudGroq
- Limits
- Free plan, gpt-oss-120b / 20b and Qwen3.8 27B: 30 RPM, 1K RPD, 8K TPM, 200K tokens/day. Whisper: 20 RPM, 2K RPD.
- Resets
- unknown reset time not stated
- Models
- gpt-oss-120b, gpt-oss-20b, Qwen3.8 27B, Whisper large v3 (+ turbo), Llama Prompt Guard 2, Orpheus TTS
- Card
- unknown the docs don't say whether the free plan needs a card.
- Training
- No · Not retained by default; abuse/reliability logs kept up to 30 days unless you turn on zero data retention.
OpenRouter :free modelsOpenRouter
- Limits
- 20 RPM and 50 requests/day; 1,000/day once you have bought $10 of credits (ever).
- Resets
- Per UTC day
- Models
- Models with IDs ending in :free (the list changes)
- Card
- No · 50/day with no purchase. The $10 credit buy lifts it to 1,000/day.
- Training
- Per provider · OpenRouter doesn't store prompts by default. Some providers may train; a separate setting for free models controls whether you are routed to them.
Workers AICloudflare
- Limits
- 10,000 Neurons a day (about 147K output tokens on gpt-oss-120b at the published rate). Text generation: 300 RPM.
- Resets
- 00:00 UTC daily
- Models
- Most of the Workers AI catalog (e.g. gpt-oss-120b)Paid plan only: Kimi K2.6 / K2.7 Code, GLM-5.2 / 5.3 / 5.3 Flash, DeepSeek V4 Flash / V4.1 Flash / V4 Pro.
- Card
- unknown the pricing page says the free allocation is on the Workers Free plan; it doesn't mention a card.
- Training
- No · Cloudflare does not train on Workers AI content; inputs and outputs are not retained after the response.
Cohere trial keyCohere
- Limits
- 20 RPM on chat models and 1,000 API calls a month. Not for production or commercial use.
- Resets
- unknown monthly reset date not stated
- Models
- Command A (+ Reasoning, Vision, Translate, A+), Command R / R+ / R7B, North Mini Code, Embed, Rerank
- Card
- No · A trial key is created with every new account.
- Training
- Yes · Trial inputs and outputs may be used for research and model improvement (de-identified first). Don't send personal data.
Gemini API (AI Studio)Google
- Limits
- unknown per-model limits are only shown in AI Studio, per project, not in the docs.
- Resets
- RPD resets at midnight Pacific
- Models
- 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-LiteNot free: Gemini 4 Argon and the image models.
- Card
- No · Free tier needs only an active project. Linking billing moves you to Tier 1.
- Training
- Yes · Free (unpaid) prompts and responses are used to improve Google products; human reviewers may read them. In the EEA, Switzerland and the UK, the paid-tier terms apply.
Mistral API (Free mode)Mistral
- Limits
- unknown per-model requests/second, tokens/minute and tokens/month are only shown in the Admin panel. Free mode has the lowest tier.
- Resets
- unknown
- Models
- unknown unknown the docs point to the pricing page and the Admin panel for which models Free mode includes.
- Card
- No · API access is on by default, no card required.
- Training
- Unclear · The help center says Free mode data may be used for training, with an opt-out in Admin > API > Privacy. The docs privacy page says API data isn't used for training; the two don't agree.
NVIDIA NIM (build.nvidia.com)NVIDIA · prototyping only
- Limits
- unknown free for NVIDIA Developer Program members, for prototyping and testing only. Rate limits aren't in the docs.
- Resets
- unknown
- Models
- NIM API catalog models
- Card
- unknown joining the Developer Program is free; card requirement not stated.
- Training
- unknown not stated in the NIM FAQ.
- Visit NVIDIA ↗
- NIM FAQ ↗
- checked
Cerebras (free trial)Cerebras · trial, not free
- Limits
- Trial only: $5 of credits that expire after 30 days. 5 RPM, 30K uncached TPM, 1M tokens/hour and 1M/day. No permanent free tier.
- Resets
- Token bucket, refills continuously
- Models
- gpt-oss-120b, Qwen 3.8 27B
- Card
- Yes · A verified payment method is required to get the $5.
- Training
- unknown not stated on the rate limits page.
Changes
GitHub
GitHub Models retired: the free playground, model catalog, inference API and BYOK are gone for all customers
Left out
- GitHub Models: Retired July 30, 2026; the playground and inference API are gone.
- Hugging Face Inference Providers: Free users get no monthly credits; PRO ($2/mo of credits) or a credit purchase is needed.
Newsletter
Get pinged when the limits move
One short email when a plan changes. No spam, we're rate-limited too.