Every call routed and checked using Jev, TypeSafe AI’s System One model

One API for every model. Always the cheapest one that can answer.

Lower bills, fewer bad answers, no subscriptions.

Live, on this page, right now

POST /v1/chat/completions

Pick a request. It goes through the same endpoint your code would call, gets triaged, answered and checked, and comes back with the receipt below.

No signup. Rate limited, because it spends real money.

What happens to one call.

Most prompts don’t need your best model.

A support classification and a schema migration go to the same model today, and you pay the migration’s rate for both. The models llm11 can reach run from $0.022 to $263 per million tokens. Most stacks pick one point on that line and send everything to it, and it is rarely the cheap end.

Routing picks the point per request. The receipt names which one it picked and what the same tokens would have cost at the top of your pool, using the figure the provider actually charged rather than a price table we maintain.

  1. llm11-fast

    $0.022 to $0.475, 8 models

    mistral-nemoqwen3-coder

  2. llm11-balanced

    $0.536 to $4.81, 8 models

    qwen3.5-27bgpt-5.2

  3. llm11-smart

    $5.63 to $263, 8 models

    gpt-5.4o1-pro

$0.022per million tokens, log scale$263

352 priced models reachable. Rates are blended 3:1 toward prompt tokens and read live from the catalogue the router ranks on, so they move when the providers move.

One line changes.

Same request body, same response body. Send model as a pack (llm11-fast, llm11-balanced, llm11-smart), pin a model id, or leave it out and use the project default.

The receipt rides along in response headers, so a parser that has never heard of us carries on working.

Your existing client

python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.openai.com/v1",
    base_url="https://www.llm11.com/v1",
    api_key=LLM11_KEY,
)

res = client.chat.completions.create(
    model="llm11-balanced",  # or a pack, a model id, or omit it
    messages=messages,
    extra_body={"context": retrieved_docs},  # optional
)

res.choices[0].message.content is unchanged.
res._llm11 carries the receipt, and response headers carry the summary.

Cheap is only safe if you check.

Routing down works when you catch the times the cheap model got it wrong. So the checking is routed too. A sub-cent triage call decides how much scrutiny an answer has earned, which is what keeps the expensive checks rare.

Drag the switches. These are the rules that run on live traffic.

The request

Short, self-contained, no source material. The common case, and the cheap one.

Criticality

Your policy sets the floor. Triage can raise it, never lower it.

Triage confidence

0.82

0.3 to 0.7 is the uncertain band.

The ladder

1 of 4 checks running

  • 00

    Schema and rules

    Free, and about a millisecond

    Valid JSON when JSON was asked for. URLs that parse. Numbers in the answer that appear in the input. Length and format honoured.

    Always runs. Deterministic, free, and never wrong about what it checked.

  • 01

    Groundedness

    Fractions of a cent

    Every factual claim in the answer checked against the context you supplied. Anything the context does not support gets flagged.

    Skipped: no context was supplied, so there is nothing to check the answer against.

  • 02

    Cross-model disagreement

    One extra completion

    A different model family answers the same question independently. Two unrelated models rarely invent the same false detail.

    Skipped: this request is not marked critical, so a second full completion is not worth its cost.

  • 03

    Self-consistency

    Three extra completions

    The same model is resampled at temperature and the spread is scored. An answer that changes between runs was never settled.

    Skipped: reserved for critical requests.

Verification spend on this request

under 1% of the ceiling

The ceiling is all four rungs on every call, which is what a verification product that checks everything would cost you. Most traffic sits at the bottom of that bar, which is the entire point.

We do not claim this removes hallucination. It catches unsupported claims against sources you supply, a second model materially disagreeing, and answers that will not hold still. What it catches, and what it misses.

Jev decides which model answers.

Jev is TypeSafe AI’s System One model. Instead of generating text, it takes the request and a typed question, “which of these models should answer this”, and returns a typed answer with a calibrated confidence in a single pass. It runs in roughly a tenth of a second for a fraction of a cent, which is the only reason deciding per request is affordable.

It is not a single point of failure. If Jev is unreachable, llm11 falls back to its own heuristic backend, and every receipt names which one decided.

Point one base URL at us and see what a day of your traffic costs.

$2.00 of credit on signup, no card. After that, credits at provider price with a 5% fee when you buy them, and nothing else.