Free Tier

Free models, daily limits, and how community-hosted free routes work alongside paid tiers.

Overview

AllRoutes offers a free tier so you can prototype, evaluate, and run small workloads without entering a credit card. The free tier has two parts:

Free credit -- $5 of one-time credit on signup that works against any model
Free routes -- a curated list of models served at zero per-token cost, subject to rate limits

Free routes are powered by community-hosted endpoints, providers' own free tiers (e.g., Groq Free, Together Free), and AllRoutes-subsidized capacity. They are best for development, low-stakes evaluation, and side projects.

What's Free

Look up free models with the Models API:

curl "https://api.allroutes.ai/v1/models?category=chat&capability=free" \
  -H "Authorization: Bearer allroutes_sk_..."

Currently free (subject to change -- always check the live list):

Model	Notes
`meta-llama/llama-3.1-70b-instruct:free`	Hosted on community vLLM
`meta-llama/llama-3.1-8b-instruct:free`	Faster, smaller variant
`mistralai/mistral-7b-instruct:free`	General-purpose, low latency
`google/gemma-2-9b-it:free`	Google's open Gemma family
`qwen/qwen-2.5-7b-instruct:free`	Strong multilingual
`nous-research/hermes-3-llama-3.1-8b:free`	Fine-tuned for chat
`microsoft/phi-4:free`	Compact, strong on math/code

The :free suffix on model IDs explicitly opts into the free route. Without the suffix, the same base model may be paid (e.g., meta-llama/llama-3.1-70b-instruct is the paid route, hitting Together / Fireworks).

Rate Limits

Free routes have stricter limits than paid:

Plan	Free-Route RPM	Free-Route Daily Quota
Free	20 RPM	200 requests/day
Pro	60 RPM	1000 requests/day
Enterprise	Custom	Custom

When a free-route limit is hit, requests return 429 Too Many Requests with error.code: "free_tier_exhausted". You can either wait for the next window or fall back to a paid route via the models array (see below).

Quotas are separate, not shared. The :free model variants have their own per-day quota (the "Free-Route Daily Quota" column above) that is tracked separately from your plan's overall RPD limit. Burning through your free-route quota does not consume your paid RPD, and vice versa.

Routing to Free First, Paid Fallback

The cleanest pattern is to set up a fallback chain:

response = client.chat.completions.create(
    models=[
        "meta-llama/llama-3.1-70b-instruct:free",
        "meta-llama/llama-3.1-70b-instruct",  # paid, same model
        "gpt-4o-mini",  # paid, different model, different provider
    ],
    messages=[{"role": "user", "content": "Hello!"}],
)

If the free route is rate-limited or down, the gateway transparently falls through to the paid version of the same model, then to a different provider. The X-AllRoutes-Model response header tells you which route actually served the request.

Free Route Strategy

You can also opt in via the routing strategy:

response = client.chat.completions.create(
    model="meta-llama/llama-3.1-70b-instruct",
    messages=[{"role": "user", "content": "Hello!"}],
    provider={"strategy": "free", "allow_fallbacks": True},
)

strategy: "free" prefers a free route for the requested model when one exists, falling through to paid if allow_fallbacks: true.

Free Tier vs. BYOK

If you have your own provider keys, BYOK is almost always a better deal than the free tier:

Feature	Free Tier	BYOK
Cost	Zero	Provider price
Rate limits	Strict (20 RPM)	Provider's limits (often 10,000+ RPM)
Model selection	~10 free models	Every model on the provider
SLA	Best-effort	Provider's SLA
Production	Not recommended	Recommended

Use the free tier for prototyping; switch to BYOK for production -- both are 0% commission.

What Free Tier Doesn't Include

The free tier covers chat completions on listed models. It excludes:

Embeddings -- embeddings are cheap (often <$0.10/1M tokens) but not free
Image generation -- always paid
Audio (TTS/STT) -- always paid
Fine-tuning -- always paid (training costs are non-trivial)
Realtime API -- always paid (high compute cost)

For experimentation in these categories, use the $5 signup credit.

Quality and Caveats

Free routes share community-hosted infrastructure, which means:

Higher p95 latency -- expect 2-5x the latency of premium providers
Occasional unavailability -- community endpoints go down for maintenance more often than paid providers
Lower throughput -- single-stream rate limits apply
No SLA -- the free tier is best-effort, not contracted

For evaluation and prototyping, this is fine. For production, route through BYOK or paid providers.

SDK Examples

Python

# Always use free first, fall through to paid
response = client.chat.completions.create(
    models=[
        "meta-llama/llama-3.1-70b-instruct:free",
        "gpt-4o-mini",
    ],
    messages=[{"role": "user", "content": "Hello!"}],
)
print(f"Served by: {response.model}")  # tells you which route was used

Node.js

const completion = await client.chat.completions.create({
  models: [
    "meta-llama/llama-3.1-70b-instruct:free",
    "gpt-4o-mini",
  ],
  messages: [{ role: "user", content: "Hello!" }],
});

cURL

curl https://api.allroutes.ai/v1/chat/completions \
  -H "Authorization: Bearer allroutes_sk_..." \
  -H "Content-Type: application/json" \
  -d '{
    "models": ["meta-llama/llama-3.1-70b-instruct:free", "gpt-4o-mini"],
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Troubleshooting

Symptom	Cause	Fix
`429: free_tier_exhausted`	Daily quota hit	Wait or use paid fallback
Free model returns gibberish	Community endpoint instability	Add a paid fallback or report on Discord
Latency 5-10x normal	Free route under load	Switch to a paid route or off-peak hours
`404: model not found`	Used `:free` suffix on a paid-only model	Check `/v1/models?capability=free` for current free list