Optim / Intelligent Multimodal Routing NEW LOOK

The cheapest model that still gets it right.

IMR prices every candidate for your prompt, picks the cheapest one above the 0.95 quality bar, shows the saving, learns from 👍/👎 and exports the answer.

Open the live demo How it routes

P(correct) ≥ 0.95 Thompson-sampling router Says which model answered Demo answer labelled as demo

run_imr·router·prompt RUNNING

Optimcheapest above barlearns from feedback

0.95

Quality barP(correct) a model must clear.

Models pricedFor every prompt, side by side.

80%

Saving, shown requestDemo: Sonnet vs default Fable.

Export formatsAnswer, decision and prices.

01 / the screen

The answer, the choice, and what the choice saved.

IMR: session totals on the left, answer cards with chosen model and saving in the middle, routing decision and priced candidates on the right — **Demo answer** — no provider key on the demo, and the label says so. With a key it shows the live provider and model.

1Where it came from — live model, or demo answer.

2The choice — model, P(correct), cost, saving.

3👍 / 👎 — retrains the router.

4Candidates — every model priced for this prompt.

5Export — PDF, Word, PowerPoint, Excel, CSV, MD, JSON.

02 / how it routes

Read the prompt, predict, pick the cheapest that clears the bar, learn.

1 · prompt

Your request

Typed, or one of the samples.

2 · features

Task & difficulty

Task, length, code, maths, difficulty.

3 · predict

P(correct) per model

Every candidate scored for this prompt.

4 · route

Cheapest above 0.95

Explores only while still unsure.

5 · learn

Feedback

👍 / 👎 updates the router; spend shows in Optim.

03 / price vs quality

The strongest model is rarely the one you need.

Exhibit A· one prompt, five candidatesblack line = 0.95 bar

ModelPredicted P(correct)Price per 1M tokens

Claude Fableeligible · default

P(correct) 0.98

$15 / 1M tokens

Claude Sonnetselected

P(correct) 0.96

$3 / 1M tokens

Gemini Flashbelow threshold

P(correct) 0.85

$0.35 / 1M tokens

GPT-5.6 Minibelow threshold

P(correct) 0.78

$0.15 / 1M tokens

GPT-5.6 Nanobelow threshold

P(correct) 0.62

$0.05 / 1M tokens

From the demo request above. Two models clear the bar; the cheaper one wins — 80% below the default.

04 / which one do I need?

Same look. Different question.

AIR Playground

“What do our documents say?”

Answers from your AI-ready internal documents
Internet only if you choose it
Citations, amendments, Witness receipts

AIR Playground

IMR

“Which model should answer, at what cost?”

Sends the prompt to a cloud model
Never reads internal documents
Routing decision, prices, savings

Optim

Next step / see it on your own documents

Route your own prompts. See the saving per request.

IMR shares Optim’s usage ledger, so every routed prompt shows up in the command center.

Book a walkthrough Back to Optim