Skip to content
← All news

News2 min read

#1 on Lean-Eval v1 first solves

Lean-Eval froze its first 128-problem release and moved everything older to an archive. On v1 our GPT-5.6 flow has the most first solves, 30, and is one behind the leader on total; on the archive it has 170 of 171.

In August this flow took first place on the Lean-Eval leaderboard. That board no longer exists in the same shape. The maintainers have frozen a first release, LeanEval v1, of 128 problems (published 20 August 2026), and moved the older problems to an Archive of 171. The leaderboard now shows one scope at a time, and a single "first place" no longer describes it. Here is where we stand in each.

301 problems · two scopes
301 problemsArchive · 171v1 · 128+2 not in v1unique · first solves · total

Swipe the diagram sideways to see all of it →

1/0 Lean-Eval has 301 problems.

All numbers below come from the board's own published data, generated 2026-10-05 12:09 UTC. Our entry is listed as "Humanifa + GPT 5.6 sol" (sic).

LeanEval v1: 128 problems, 88 solved by anyone ​

First solves · #1

30

The first accepted proof on 30 of the 88 problems anybody has solved.

Total · #2

79

One behind NEAR AI’s 80.

Unique · #4

1

One of our 79 has no other solver.

The board sorts by unique by default, so that is the position a visitor sees first. Sort the table below by any column to see the board each way.

LeanEval v1 · 128 problems · click a column to sort
Humanfia, GPT-5.613079
Axiom Prover (Axiom Math)32174
NEAR AI, with DeepSeek V431280
Humanize2 flame chase: GPT-5.6-sol + Claude Opus 5077
max gp (GPT-6)2217
Tau (caj.al)0029
Midas0028
  • First solves: first, with 30. We were the first to have an accepted proof on 30 of the 88 problems anybody has solved.
  • Total: second, with 79, one behind NEAR AI's 80.
  • Unique: fourth. Only one of our 79 has no other solver.

The three columns reward different things: being first, being alone, or being broad. We lead on the first and are close on the third. Others have more problems nobody else can do.

Archive: 170 of 171 ​

On the 171 archived problems we have accepted proofs for 170, tied for the most with two other entries.

Archive · 171 problems · accepted proofs
  1. 1Axiom Prover (Axiom Math)170
  2. 1Tau (caj.al)170
  3. 1Humanfia, GPT-5.6170
  4. 4Seed Prover (ByteDance)155
  5. 5Aristotle (Harmonic)150
  6. 6GPT-5.6112
Ties share a rank. Counted from the board's published data.

249 in all ​

Counted across both scopes, 249 distinct problems carry an accepted Humanfia proof, the most of any entry on the board.That count is ours, made from the published data; the site does not show a combined column. The board also notes that one model may appear under more than one name while names are consolidated. Read by acceptance date, the lead has changed hands: Axiom Prover's first 206 proofs were all accepted on 29 August and put it ahead, and our September and October proofs put us back in front on 4 October.

Lean-Eval · all scopes · accepted proofs per entry

To 249

Swipe the plot sideways to see all of it →

Submitted by Zhengyang Zhang

210

Of the 249 accepted solutions.

By Hongzhou Lin

34

And 5 by Jui-Hui Chung.

Since 19 August

75

Accepted on or after the day after our first place.

What counts ​

Lean-Eval accepts a solution only when it passes Comparator against the problem's statement. There are no partial credits and no sorry.

The leaderboard · its source · HOA

Read next

News30/30 IPhO, 6/6 IOI, 100/100 IBOIPhO 2026 theory 30 of 30, IOI 2026 six of six, IBO 2024 theory 100 of 100, IChO 2026 formalized 68 of 68, and quantum information theory 40 of 40. What each score is, who graded it, and what it does not show.Read it News672/672 on PutnamBench, joint #1GPT-5.6 in a worker–reviewer loop proved all 672 PutnamBench statements in Lean, at $44.50 a problem. The top of the official leaderboard is now a four-way tie, and what separates the four is cost.Read it News#1 on Lean-Eval: 172 proofsThree weeks after second place, a fully agentic GPT-5.6 run is first on the Lean-Eval leaderboard, with 172 research-level problems solved. Every accepted proof is sorry-free and independently re-verified.Read it