2026 Verified Developer Directory · Zero Login · OpenAI SDK Compatible

Free LLM API Providers & Free AI Token Quotas (2026)

Every active free LLM API key and starter credit tier in one filterable table—with exact RPM / RPD / TPM rate limits, credit card requirements, prompt training policies, and 2026 shutdown alerts (including GitHub Models & Hugging Face policy changes).

Active Free LLM APIs
9 Verified Providers
No Credit Card Tiers
8 Instant API Keys
2026 Retired / Removed
2 Flagged (Avoid 404s)
Last Full Audit
Nov 19, 2026
Enter your expected daily requests (e.g. 200 or 600) to filter out low-cap tiers. 11 of 11 providers shown
Provider & Top Free Models Exact Free Quota (RPM / RPD / Credits) Card & Training Policy The Catch / Hidden Rules Last Verified Quickstart
Google AI Studio (Gemini API)
Permanent Free Tier
gemini-2.5-flashgemini-2.5-progemini-3.5-flash-lite
10 RPM · 500 RPD · 250K TPM (Flash) / 2–5 RPM · 25–100 RPD (Pro)
Permanent Daily Quota (Resets Midnight PT)
No Credit CardTrains on Free Tier
Used for training on Free Tier (Unpaid service in supported regions)
Rate limits apply per Google Cloud project (not per API key). Free tier inputs/outputs may be reviewed for product improvement unless Billing is linked.
2026-11-19
Get Free Key ↗
Groq Cloud
Permanent Free Tier
openai/gpt-oss-120bqwen-qwq-32bmeta-llama/llama-prompt-guard-2-86m
30 RPM · 1,000 RPD · 8K TPM · 200K TPD (Open models)
Permanent Daily Quota
No Credit CardNo Training
No training on API prompts by default
Standard production Llama 3.1/3.3 models moved off self-serve free tier; open-weight models (gpt-oss, Qwen) remain on the 30 RPM / 1,000 RPD free baseline.
2026-11-19
Get Free Key ↗
OpenRouter (:free Models)
Permanent Free Tier
deepseek/deepseek-r1:freegoogle/gemini-2.5-flash:freeqwen/qwen3-235b-a22b:free
20 RPM · 50 RPD ($0 balance) or 1,000 RPD (one-time $10 lifetime top-up)
Permanent Daily Quota across :free models
No Credit CardTrains on Free Tier
Varies by upstream provider (Many :free providers log prompts)
Global cap across all `:free` models combined. A one-time $10 credit purchase permanently unlocks 1,000 RPD on `:free` models without spending the $10.
2026-11-19
Get Free Key ↗
Cloudflare Workers AI
Permanent Free Tier
@cf/meta/llama-3.3-70b-instruct-fp8-fast@cf/qwen/qwen2.5-coder-32b-instruct@cf/deepseek-ai/deepseek-r1-distill-qwen-32b
10,000 Neurons / day (Resets 00:00 UTC, ~150–400 LLM calls/day)
Permanent Daily Compute Quota
No Credit CardNo Training
No customer data used to train models
Measured in 'Neurons' rather than fixed request counts; larger 70B models consume Neurons faster than 8B models.
2026-11-19
Get Free Key ↗
Mistral AI (La Plateforme Experiment)
Permanent Free TierPhone SMS
mistral-small-latestcodestral-latestpixtral-12b-2409
1 RPS (~60 RPM) · Evaluation TPM/Monthly Token Pool
Permanent Experiment / Prototyping Plan
No Credit CardTrains on Free Tier
Used for model training on Free Experiment tier
Requires phone number verification. Strictly throttled at 1 request/sec (1 RPS) and prompts are opted into model training unless upgraded to paid tier.
2026-11-19
Get Free Key ↗
Cohere (Trial API Key)
Permanent Free Tier
command-a-03-2025command-r-plus-08-2024rerank-v3.5
20 RPM (Chat) · 1,000 API calls / month total
Permanent Monthly Trial Quota
No Credit CardTrains on Free Tier
Trial usage restricted to non-commercial evaluation
Hard cap of 1,000 total API calls per month across all endpoints combined (~33/day average); strictly non-commercial.
2026-11-19
Get Free Key ↗
SambaNova Cloud (Free Tier)
Permanent Free Tier
Meta-Llama-3.3-70B-InstructDeepSeek-V3.1
20 RPM · 20 RPD · 200,000 TPD (No card linked)
Permanent Daily Quota
No Credit CardNo Training
Zero data retention on inference API
Without a linked payment card, daily requests are capped at 20 RPD per model (200K TPD). Linking a card unlocks Developer Tier (up to 48,000 RPD).
2026-11-19
Get Free Key ↗
NVIDIA NIM (build.nvidia.com)
Starter CreditsPhone SMS
deepseek-ai/deepseek-r1meta/llama-3.3-70b-instructnvidia/llama-3.1-nemotron-70b-instruct
1,000 Starter Credits · ~40 RPM prototyping cap
One-Time Starter Credits
No Credit CardNo Training
No training on API payloads
One-time 1,000 credits pool for prototyping; hard 40 RPM system cap on free accounts cannot be raised via forum requests.
2026-11-19
Get Free Key ↗
Cerebras Inference ($5 Free Trial)
Starter Credits
gpt-oss-120bqwen-3.8-27b
$5 Free Credit (Expires in 30 days) · 5 RPM · 1,000,000 TPD
30-Day $5 Trial Credit (Card Required)
Card RequiredNo Training
No model training on API traffic
No longer offers a permanently card-free tier; requires adding a verified payment method to claim the 30-day $5 trial credit, capped at 5 RPM.
2026-11-19
Get Free Key ↗
GitHub Models (Retired Jul 2026)
Retired in 2026
Moved to Azure AI Foundry / GitHub Copilot
RETIRED on July 30, 2026 (0 RPM / Shut Down)
Discontinued Service
DiscontinuedN/A
Service Discontinued
Official GitHub Models playground, catalog, and free inference API were permanently retired on July 30, 2026. Most other directories still list it by mistake.
2026-11-19
Hugging Face Inference Providers (Free Credits Removed)
Retired in 2026
Requires Prepay or PRO ($9/mo gets $2/mo credits)
$0.00 / mo for Free Users ($0.10/mo free credits removed)
Free Credits Discontinued
DiscontinuedN/A
Varies by routed provider
Hugging Face removed the $0.10/month free inference credit allocation for free accounts due to automated account abuse. Requires prepaid balance or PRO plan.
2026-11-19

⚠️ 2026 Deprecation Watch: What Stopped Being Free?

  • GitHub Models (Shut Down July 30, 2026): Most older blogs still recommend models.inference.ai.azure.com, but GitHub officially retired the free GitHub Models playground and inference API on Jul 30, 2026.
  • Hugging Face Serverless Inference ($0.10/mo Removed): Free Hugging Face accounts no longer receive monthly free inference credits (`None`) due to automated abuse; PRO ($9/mo) or prepaid balance is now required.
  • Cerebras Inference: Shifted from an open card-free tier to a $5 30-day free trial credit that requires linking a verified payment method (capped at 5 RPM).
  • Groq Llama 3.3 70B Shift: Standard production Llama 3.1/3.3 models transitioned toward enterprise/paid access, while open models like openai/gpt-oss-120b and qwen-qwq-32b remain on the 30 RPM / 1,000 RPD free baseline.

🔑 How to Combine Free LLM API Keys Without Hitting 429

  • Primary High-Context Workhorse: Start with Google AI Studio (gemini-2.5-flash gives 10 RPM / 500 RPD / 250K TPM with no credit card).
  • High-Speed Burst Fallback: Add Groq Cloud (30 RPM / 1,000 RPD on open models) and NVIDIA NIM (1,000 starter credits at 40 RPM).
  • OpenRouter $10 Unlock Trick: OpenRouter caps :free models at 50 RPD on $0 accounts, but a one-time $10 credit top-up permanently unlocks 1,000 RPD across all :free models without spending your $10 balance.
  • Using OpenCode or Claude Code? Jump to our OpenCode Free Models & JSON Config Generator to export a multi-provider fallback config in one click.

🚀 Hit the 429 Rate Limit or Need Zero-Data-Training for Production?

Free API tiers are great for prototypes, but most either train on your prompts (Google AI Studio Free, Mistral Free, OpenRouter :free) or cap out at 5–30 RPM. When your app or Coding Agent outgrows free limits, these are the lowest-cost upgrade paths: