Open models. One honest gateway.
Every open model. One key. The best price, and it only moves down.
- lineup
- 143 models
- verified, 7d
- 0 of 0 models
- cheapest, mistral-nemo-instruct-2407
- $0.03 / $0.03 per 1M
- uptime, 24h
- 95.7%
The same open model is not the same everywhere
price spread for the same open model across providers
benchmark score of identical weights, depending on who serves them
open models across every provider we route to, behind one key
What honest means here
Every claim below has a public endpoint behind it. These blocks are reading them right now, failures included.
Every route is tested before it serves you
The same weights differ by host, so canaries test tool calling, code generation and unicode on every route. A route that fails never serves your tool calls.
0 of 0 models pass every canary over the last 7 days
Failover you never notice
Requests go to the cheapest healthy provider. When one degrades, the next one answers the same request, and your code sees one reply.
Measured over the last 24 hours
Prices only move down
Every price tracks the cheapest provider. A provider cut passes through automatically, and a mid-month cut re-rates your whole month as a rebate.
Price history8 automatic cuts passed through, most recent first
Drop-in by design
Works with any OpenAI or Anthropic SDK and the agents built on them: Claude Code, Codex, Cline, OpenCode. Point the base URL at Infero and keep everything else.
from openai import OpenAI
client = OpenAI(base_url="https://api.infero.xyz/v1", api_key="sk-infero-...")
r = client.chat.completions.create(
model="deepseek-v4-flash",
messages=[{"role": "user", "content": "Hello!"}],
)
print(r.choices[0].message.content)Fair with money
Prepaid only
You spend what you have put in, and a surprise bill is impossible.
Credits never expire
Your balance stays yours until you spend it.
1:1 crypto top-ups
On Base, at the signed rate. Pay in ELSA and get 5% extra credit, up to $25 per account.
USDC
ELSA+5% extra credit, up to $25 per accountPer-key budgets
Caps and model allowlists give a leaked key a blast radius.
Questions
- What is Infero?
- An inference gateway for open models: DeepSeek, Qwen, Kimi, GLM and the rest, for chat as well as image, speech and transcription. One API key, compatible with the OpenAI and Anthropic SDKs, routed across several upstream providers with automatic failover.
- How is pricing set?
- Every model's price tracks the cheapest provider, per million tokens. Prices only move down automatically. When a provider cuts its price the sticker follows, and the cut re-rates your current month as a rebate. The full price history is public.
- What does verified serving mean?
- The same open weights can behave differently depending on who hosts them. Every model and provider pair is tested with canaries for tool-call conformance, code generation and unicode round-trips. Scores are public, and requests that use tools never route to an endpoint failing the tool-call canary.
- How do I pay?
- Prepaid credits that never expire. Signing in creates the account, with $0.50 of free credit when you sign in with Google; top up with USDC or ELSA on Base, credited 1:1 at the signed rate, gasless. Paying in ELSA adds 5% extra credit, up to $25 per account.
- Does it work with Claude Code, Codex and other agents?
- Yes. Anything that speaks the OpenAI or Anthropic API works by changing the base URL: set OPENAI_BASE_URL or ANTHROPIC_BASE_URL to the gateway and keep the rest of your setup.
Try it on us
$0.50 of free credit when you first sign in with Google. No card, one account per person.
Start free