Free tier
Real models, free, without a card
25 requests a day on real production models - not a sandbox, not a countdown. It refreshes daily and never expires, because a trial that runs out while you are still deciding tells you nothing about the product.
What the allowance gives you
- 25 requests every day, reset daily and never expiring
- Answers up to 8,000 tokens
- No credit card, and nothing charged if you never add one
- The same routing, fallback, guardrails and receipts paid traffic gets
The models it covers
Two, named plainly rather than described as “selected models”. Sending "auto" picks the first.
openai/gpt-5-miniGPT-5 mini
400,000-token context · $0.25 in · $2 out once you are paying.
deepseek/deepseek-v4-flashDeepSeek V4 Flash
1,000,000-token context · $0.22 in · $0.66 out once you are paying.
One caveat worth stating: the second model's provider trains on prompts by its own published policy, which is why "auto" never chooses it - you have to name it. Every model's data policy is on the model directory with the date we checked it.
Start now
curl https://finalrouter.com/api/v1/chat/completions -H "Authorization: Bearer $FINAL_ROUTER_KEY" -H "Content-Type: application/json" -d '{
"model": "auto",
"messages": [{ "role": "user", "content": "Hello" }]
}'
# Every response says how much allowance is left:
# X-Final-Router-Free-Remaining: 24What the allowance does not cover
Said here rather than discovered in an error. The free allowance is a chat allowance: embeddings need credit, because none of its three bounds mean anything for vector workloads. Routing policies, the rest of the catalogue and higher rate limits also arrive with credit - and credit is pay-as-you-go, with no subscription and no minimum.