Skip to content
Several seats · one answer01/07

SEVERAL OF THE
BEST MODELS.
ONE ANSWER.

One request reaches every seat in turn. Each takes the part it leads on and leaves the work better than it found it. What comes back is one answer, with the reasoning underneath it.

See what it costsSee a real run
Three seats on StarterAll five from ProFrom $29/mo · card required
Codewrites it
Researchsources it
Reasoningargues it
Creativenames it
Visionlooks at it
One answer
A real run · 8 September 2026
PARTICLES ON A SPHERE

The Research seat refused to sign it. Its objection is on the record, the others did not adopt it, and you get both.

2 signed · 1 refused · 2 silent — see the run
02/07How it works

A RELAY, NOT A VOTE.

Five models answering the same question in parallel and averaging the result produces the blandest of the five. So they do not work in parallel. Each seat receives the work so far and improves it in the one dimension it leads on — and may rewrite anything earlier if it can genuinely make it better.

The hand-off carries the reasoning, not just the conclusion. That is the mechanism by which several models beat one: each seat inherits the thinking of the seat before it.

Occasionally a seat will not put its name on the result. When that happens it is recorded and shown beneath the answer, rather than averaged away. Most runs have no disagreement, and that is the point — one that always disagrees means nothing.

A real run

· BUILD · 470s

What was asked

Build an R3F scene: 2000 instanced particles on a sphere that push away from the pointer and settle back. React Three Fiber v9, drei, TypeScript.

What each seat got, and what it passed on

Read down. Each seat was handed the reasoning of the one above it — not just its conclusion — and this is what it changed.

  1. Researchhanded the requestBuilt the interaction with one instanced draw call, preallocated simulation buffers, direccut at 90
  2. Reasoninghanded research findingsReplaced the four independent physics coefficients with two stated design intents — hole dcut at 90
  3. Codehanded reasoning findingsReplaced the frame-rate-dependent, home-position-sampled force loop with a fixed 1/60 s accut at 90
  4. Visionhanded code findingsOptimized the particle update path with dynamic instance-buffer usage, squared-distance tecut at 90
  5. Creativehanded vision findingsThe “after” version is truncated mid-line in the package.json script block, cutting off “pcut at 90

Who signed it

  • Reasoning — signed (no answer — treated as a sign-off)
  • Code — signed
  • Research — refused
  • Vision — none recorded
  • Creative — none recorded

The Research seat refused to sign it

“Do not ship or endorse this as it stands. The `"preview"` script is cut off at `"v`, and the JSON is never completed or closed, so package managers cannot parse the file and none of its scripts can reliably run. Replace it with the complete intended `package.json`—including a valid preview command, closing syntax, and the project’s dependencies—then run `npm install` and `npm run build` before trecut at 400

The piece optimises for presenting an ostensibly finished project despite containing a truncated, invalid `package.json`, which costs this person a buildable or reviewable result.

On the record. The others did not adopt it.

Two of the five saw it. The last row above is the Creative seat reporting the same broken file in its hand-off, and Research then refused to sign over it. The answer shipped anyway — and so did this. Seeing the objection at the same moment as the answer is the thing you are buying.

The run log stores the first 90 characters of each hand-off and the first 400 of the dissent, so both stop mid-word. That is where the record ends; nothing has been completed or tidied. The workspace it belongs to, and the model behind each seat, are the only things withheld. Run 2ce1ec8c-34f1-4f53-98c0-c84a86ad036b.

03/07The seats

FIVE CAPABILITIES. ONE LEAD EACH.

Shown by capability, not by model name: the roster changes as models are released, and a name we have not verified is a name we will not print. This is the roster, not a running order — the council runs in whatever order the request’s shape calls for, and the run above was a BUILD.

01
Seat 01Code

IT WRITES THE THING

Large refactors, type discipline, builds that pass first time. When the answer has to run, this is the seat that makes it run.

02
Seat 02Research

IT FINDS THE SOURCE

Retrieval, citation, and then the source behind the source. It comes back with what the others would have asserted from memory.

03
Seat 03Reasoning

IT HOLDS THE ARGUMENT

Long chains, contested premises, trade-offs with no clean answer. It is the seat that refuses the shortcut the other four would take.

04
Seat 04Creative

IT KILLS THE SAFE OPTION

Naming, tone, direction. The part a committee averages into nothing, held by the one seat willing to be disagreed with.

05
Seat 05Vision

IT READS THE ROOM

Screenshots, whiteboards, scanned pages, the photograph of a napkin. The seat that can look at what you actually have.

04/07What came back

Pointer-repelled instanced particles on a sphere (R3F v9)

2000 particles rendered as a single InstancedMesh, distributed evenly on a sphere with a Fibonacci lattice. The pointer is raycast against an invisible sphere of the same radius; particles inside the influence radius of the hit point are pushed outward along the surface, and a damped spring pulls every particle back to its rest position. All simulation state lives in typed arrays and is written into instance matrices each frame, so nothing allocates in the hot loop and React never re-renders during interaction.

And what it refused to pretend: It deliberately does not use a GPU compute pass, does not react to touch pressure or multiple pointers, and does not simulate inter-particle collision — each particle only knows its rest point and the pointer.

See what a run costsA card before the first run
05/07Your work

WHAT YOU PASTE IN STAYS YOURS.

Three things about your work, each of which the architecture actually enforces — not a policy we ask you to take on trust.

01

Your request goes to several model providers

Each one is named on the sub-processor list, with what it receives and which country it processes it in. We make no claim about whether they train on it — their terms differ per model and change without notice, so it is not a promise we could keep.

02

Every run is traceable

You see which model wrote which part, what it was given, and where it would not sign off. Nothing is merged silently.

03

Delete a run and it is gone

The row and everything attached to it are removed. There is no hidden copy marked deleted.

Every service that receives your data is named, with what it gets and where it runs.Read the sub-processor list
06/07Pricing

THE FRONTIER COUNCIL STARTS AT PRO.

Plans do not run the same models. Yeta Starter runs three standard seats. Yeta Pro and above run the full five-seat frontier council. Starter is not worse models — it is three good ones. The frontier seats are what Pro buys.

Yeta — one person

Yeta Starter

$29
per month
In India: ₹2,900 incl. GST
3 seats on every run, every month — and the plan every other plan is measured against.

One person, three standard seats on every run.

  • 3 seats — the standard council
  • Full trace and dissents
See Yeta Starter

Yeta Pro

$79
per month
In India: ₹7,900 incl. GST
3x
the usage of Yeta Starter

The frontier council, for one person.

  • 5 seats — the full frontier council
  • Full trace and dissents
See Yeta Pro

Yeta Max

$199
per month
In India: ₹19,900 incl. GST
8x
the usage of Yeta Starter

The frontier council at depth, for one person.

  • 5 seats — the full frontier council
  • UltraZeta depth
  • Full trace and dissents
See Yeta Max

Zeta — a team

Zeta Starter

$99
per month
In India: ₹9,900 incl. GST
3.5x
the usage of Yeta Starter

The frontier council, shared by a small team.

  • 5 seats — the full frontier council
  • Built for a team
  • Full trace and dissents
See Zeta Starter

Zeta Pro

Recommended
$299
per month
In India: ₹29,900 incl. GST
10x
the usage of Yeta Starter

The frontier council at depth, for a team that runs all day.

  • 5 seats — the full frontier council
  • UltraZeta depth
  • Built for a team
  • Full trace and dissents
See Zeta Pro

Zeta Max

$799
per month
In India: ₹79,900 incl. GST
20x
the usage of Yeta Starter

A department.

  • 5 seats — the full frontier council
  • UltraZeta depth
  • Built for a team
  • Full trace and dissents
See Zeta Max

Prices in US dollars, billed monthly. Every plan starts on a short paid trial of that same plan. Unused capacity does not carry over — the refunds page says exactly what happens when you cancel.

07/07Start

YOU SEE WHERE
THEY DISAGREED.

Every run, on the record.

  • No password to stealYou sign in with a link sent to your email. There is no password of yours for anyone to take from us.
  • A card before the first runEvery plan is paid, so a card is required at signup and the first payment is taken before you can run anything. Nothing here is free.
  • Cancel from the account pageCancelling stops the next payment. You keep the period you have paid for.
See the plans