Introducing Quasar 1.0 – our strongest model

Convergent Discovery

Convergent puts frontier AI models on open problems in math and science, and publishes only what can be checked.

Convergent Discovery

Frontier models on open problems.

We use strong AI models such as Fable 5.1, Opus 5.5, Kimi K3 and Astra 6 to attack unsolved problems in math and science, and to find new things. Every result is checked before we call it one.

How it works

Models propose, models critique, code decides.

  1. Pick a problem that has a clear, checkable answer.
  2. Propose. One model works on a solution, a proof or a construction.
  3. Critique. Other models look for errors and gaps.
  4. Verify. Code, simulations or a proof checker decide whether the result holds.
  5. Publish the problem, the method, the result, its status and the code, so anyone can check it.

Every result has a status

We never present a result as more certain than it is.

Status 1

Unverified

A model found it. Nobody has checked it yet.

Status 2

Numerically checked

Code or simulation confirms it for the tested cases.

Status 3

Formally proven

A proof checker such as Lean accepts the proof.

Status 4

Confirmed by experts

Independent researchers have reviewed and confirmed it.

Math

Parker Square: search for a 3×3 magic square made of nine squares, made by Fable 5.1 (High). No solution up to centre 1,000,000².

Status: Numerically checked

Physics

No projects yet.

Quantum Physics

No projects yet.

Space Physics

No projects yet.

Science

Aerozith: design of a new lightweight aluminium-lithium alloy, made by Fable 5.1.

Status: Unverified

Projects

2 projects so far

  • Parker Square (Math): search for a magic square of squares. Status: Numerically checked.
  • Aerozith (Science): design of a new lightweight alloy. Status: Unverified.