Every frontier model in one place, for less.

One subscription from $5. A router picks the right model for each message and shows you why. Bring your own models, and they never count against your plan.

One month of Pro free. No credit card required.

auto → haiku 4.5 · quick lookup

Opus 4.8GPT-5.6Gemini 3.1 ProGrok 4.3Sonnet 5Haiku 4.5GPT-5.4 miniFlash Lite+ yours

The router

The right model, every message.

Quick question? A fast model answers in a blink. Hard problem? The heavyweight steps in. You can always pick a model yourself, but you rarely need to.

auto → haiku 4.5 · quick lookup

Judgment you can read

Every routed message shows which model answered and why, right on the message. The picking is visible, not magic.

auto → opus 4.8 · deep reasoning

Quality where it counts

Hard reasoning goes to frontier models. Routine work goes to fast ones. The bar for every answer stays the same.

auto stretches your allowance

Cheaper by construction

Sending everything to the most expensive model is how subscriptions get expensive. Routing is how ours stays cheap.

Your models

Every model, one place. Including yours.

The included catalog covers the frontier. Connections cover everything else: the model running on your own machine, your company's cloud, or an account you already pay for. Connected models sit in the same menu, work with the same router, and are never metered by us.

included

The frontier, included

Claude, GPT, Gemini and Grok model families in one catalog, running on our accounts, covered by your plan.

connection · your machine

Local models, first class

Point Heuricity at a model running on your own hardware and it becomes a citizen of the same chat, router and all.

connection · your account

Your accounts, your rates

Connect an OpenAI, Mistral, DeepSeek or OpenRouter account and pay the provider directly. We never count it against your plan.

Deeper technical detail on connections lives on the models page.

Pricing

Plans from $5. The math is honest.

Most single-model subscriptions cost more than all of Heuricity. Usage is measured in what things actually cost, the app shows you what you saved, and your own connections never touch your allowance.

Full detail, including exactly how usage works, on the pricing page.

When it matters

For the decisions you cannot get wrong.

Put two to four models side by side and read the differences. Or convene a council: the models argue it out and you get one answer, with the disagreement shown rather than smoothed over.

where they split

gpt-5.4

Clear the quality floor before a broad launch. Roughness spends trust.

opus 4.8

Ship past your best guess and watch where tolerance breaks. That is signal.

Agreed: rough, contained rollouts are right for finding the wedge. Irreversible failures need a hard quality gate.

Privacy

Private, and specific about it.

No tool at Heuricity shows anyone your chats. Your content never trains anything. Deleting really deletes. The privacy page says exactly who can see what, which is more than most will tell you.

Read the privacy page

Run it inside your company.

We build custom Heuricity deployments that run on your infrastructure, with your models and your data.

Talk to us