Beta
Pricing · Free TierLive · 16 free APIs · 3 kinds of free · Updated weekly

Free AI Models · Tracked

Every free way to run AI in production · from free-forever open weights to quota-gated hosted APIs and BYOK-powered IDEs. What actually costs nothing, and what the fine print says when you read past the landing page.

What does "free" actually mean?
Free forever

Open weights · you run the model on your own hardware. Inference cost is your electricity. No rate limits, no data sharing, no provider ever.

Free with quota

Provider gives you a fixed daily or monthly request budget. Great for dev and prototyping. Usually trains on your inputs · check the fine print.

Free with BYOK

The software is free · you bring your own API key. Cursor, Cline, Aider, Continue. You still pay the underlying provider bill, but tooling is zero.

6 open-weight models

Self-hostable. Apache 2.0 or permissive licenses. Zero ongoing API cost · the only cost is the GPU you rent or buy to serve them.

Self-host unlimited
Requires GPU hardware · free weights under Apache 2.0
codingreasoninggeneral
Go to provider
Self-host unlimited
Runs on consumer GPUs · Apache 2.0
codinggeneral
Go to provider
Self-host unlimited
Llama Community License · free for under 700M MAU
codingreasoningvision
Go to provider
Self-host unlimited
MIT license · free weights on HuggingFace
codingreasoning
Go to provider
Self-host unlimited
Apache 2.0 · multilingual
codingreasoningmultilingual
Go to provider
Self-host unlimited
Apache 2.0
codinggeneral
Go to provider
6 quota-gated free tiers

Fully hosted · no infrastructure. Rate-limited per day or per minute. Great for dev, risky for prod.

Google AI Studio logo
Gemini 3 Flash
Google AI Studio
15 req/min · 1.5K req/day
Google trains on your inputs in the free tier
codingvisiongeneral
Go to provider
Google AI Studio logo
Gemini 3 Pro
Google AI Studio
2 req/min · 50 req/day
Free tier logs prompts for training
reasoningvision
Go to provider
30 req/min · 14.4K req/day
Free dev tier · rate-limited hard
codingreasoningspeed
Go to provider
$5 credit on signup
One-time credit · ~500K tokens of V3.2
codingreasoning
Go to provider
$150/mo credits via xAI console
Requires sharing API usage data · enterprise only
reasoninggeneral
Go to provider
$5 free credits on signup
One-time · expires in 30 days
codingwriting
Go to provider
4 BYOK tools

Software is free. You pay the underlying API. Best for heavy users who already have keys.

Free models (zero-cost route)
Rotating catalog · rate-limited · data may be used for training
codinggeneral
Go to provider
2K completions · 50 slow premium/mo
Bring any API key for unlimited use
codingide
Go to provider
C(
Cline
Cline (VS Code)
Unlimited (BYOK)
Open source · costs depend on your API key
codingagents
Go to provider
A(
Aider
Aider (CLI)
Unlimited (BYOK)
Open source CLI · uses any LLM API
codingcli
Go to provider
Quick pick
Use caseFree foreverFree with quotaBYOK tool
Codinggpt-oss-120b · gpt-oss-20b · Llama 4Gemini 3 Flash · Llama 4 Maverick (Groq) · DeepSeek V3.2 (Together)OpenRouter free models · Cursor free plan · Cline · Aider
Reasoninggpt-oss-120b · Llama 4 · DeepSeek V3.2Gemini 3 Pro · Llama 4 Maverick (Groq) · DeepSeek V3.2 (Together)
VisionLlama 4Gemini 3 Flash · Gemini 3 Pro
Generalgpt-oss-120b · gpt-oss-20b · Mistral Small 3Gemini 3 Flash · Grok 4OpenRouter free models
MultilingualQwen3.5 397B-A17B
  • Training on your inputs · Google AI Studio, Groq free tier, and most $5-credit promos train or log prompts. Paid tiers usually have zero-retention options.
  • Rate limits that kill prod · 1500 req/day on Gemini free tier sounds plenty · until you spike. Most free APIs do not have overflow pricing.
  • Context cap downgrades · Free tiers usually cap context smaller than paid (e.g., 32K vs 1M). Check before designing your app around the free tier.
  • Geographic restrictions · Grok free, Alibaba Qwen direct, and DeepSeek free often have region lockouts or KYC requirements.
  • "Free" that is really trial credits · $5 signup credit lasts 300K-700K tokens and then ends. Plan the migration path before you commit.
For hosted zero-cost: Gemini 3 Flash (1.5K req/day free), Groq free tier (30 req/min on Llama 4), Together AI ($5 signup credit). For unlimited use: run any open-weight model like Llama 4, Qwen3.5, DeepSeek V3.2, or gpt-oss on your own hardware.