Mission
The democratisation of all available intelligence.
The best thinking machines ever built are being rented out by the handful of companies that own them, one brand at a time, to whoever can pay each of those companies separately. That is not how a general-purpose technology is supposed to reach people. Keen Labs exists to put allof it — every lab's models, and the live tools no model carries — behind one seat, at one price, chosen by evidence instead of by whose logo you happened to trust.
What “all available” means
There is no single best model, and there never was — there is a model that is best at long documents, another that is best at code at a tenth of the price, and a question whose honest answer is a live lookup rather than any model at all. Choosing between them by reputation is guesswork. So the lab measures them: 335 models reachable and measured by our catalogue pipeline (counted, not rounded, as of 2026-08-08), with 12+ in active rotation at any time.
Democratisation here is not a discount. It is the removal of an expertise tax: today, getting good answers cheaply requires knowing which model to ask, what it costs, and when to stop paying frontier prices for a greeting. That knowledge is the privilege we are trying to delete. The router carries it so you do not have to, and the receipt under every answer shows you what it decided — see the method.
The honesty covenant
An AI company's only real asset is whether you can believe its numbers.
This lab began by catching its own autonomous workforce reporting success it had not earned, and publishing the audit instead of quietly patching it. Six rules came out of that, and they are binding on everything we publish. Each one is checkable against this site rather than against our good intentions.
- 01
We never assert super-intelligence.
It is a word the measurements may one day earn, and until they do, it is marketing. We measure toward it and publish where we are. What we assert today, and prove with rows, is the democratisation itself — not the arrival of something smarter than you.
- 02
No claim without a record.
Every factual statement on this site resolves to an entry with a source and a date. If a figure cannot be cited, it does not ship — and a number with no source is treated as a defect, not a rounding error.
- 03
A sample says it is a sample.
The routing panels, the memory maps and the receipt cards are scripted replays. Each one says so on the element itself, at the same visual weight as the thing it is demonstrating — not in grey text at the bottom of the page.
- 04
Illustrative is not measured.
Figures derived from public list prices, or from arithmetic rather than observation, are labelled as such next to the number. When a measurement replaces one, we say which it was and what changed.
- 05
Nothing pulses live unless it is live.
A live indicator on this site is wired to something live. If the thing behind it is paused — as our autonomous workforce currently is — the page says paused, in the present tense, rather than animating a status we are not in.
- 06
The limits sit next to the claim.
Every argument we make carries its own cost of admission in the same room: the product page has an honest-limits section, the comparison table has rows that go against us, and the audit is the least flattering document we publish.
The proof that these are rules and not slogans is the audit: every dated, sourced claim on this site, including the ones that cost us something to publish.
Constraints, not positioning
€49 and neutrality are engineering requirements. They just look like marketing.
A mission that costs the company nothing is a slogan. These four commitments each remove an option we would otherwise have, and each one of them lands on the codebase.
One price, €49 a month, that an ordinary person can actually pay.
What it bindsA flat consumer price with frontier models inside it only works if most work never reaches a frontier model. That makes routing quality a solvency problem, not a feature: if the router sends everyday prompts down expensive lanes, the price has to rise or the promise breaks. The engineering is not allowed to be lazy here, because the business model has no slack for it.
Neutrality — we sell no model of our own.
What it bindsA lab that sells its own model cannot afford to tell you when a rival's is better. We buy every model, so honest ranking is the business rather than a conflict with it. The constraint this imposes: no house model, no favoured vendor, no ranking a model can pay to change. That is not a claim about anyone's engineering — it is business-model logic, and it is checkable.
A receipt under every answer.
What it bindsModel, latency and price, visible on the answer itself. This forces per-answer metering rather than a monthly guess, and it makes a silent downgrade impossible to hide: if we quietly routed you somewhere cheaper, the receipt would name it.
Findings public, routing configuration private.
What it bindsWe publish the method and what we found; we do not publish the live routing rules. This is a deliberate limit on how far you can verify us from outside, and stating it plainly is better than implying an openness we do not offer. The findings are the pitch; the config is the moat.
Premium is €49/month, founder-reported, verified 2026-05-21. The arithmetic behind what that buys — and the fact that it is arithmetic on public list prices rather than a measured average — is shown in full on /pricing.
What we don't claim
The mission is also a list of things we refuse to say.
Hold us to it
A covenant nobody can check is a brochure.
If a number on this site has no source, a sample is not labelled a sample, or a claim outruns its evidence, that is a defect — email hello@keenlabs.pro and it gets corrected in public. Pilot access is how you check the product itself: one month of Premium, free, with a direct line to the founder.