Assertion Contradicted AI assessment confidence: 95% certainty 4/5 debate potential 2/5

The open-source Lean dataset contains only tens of millions of tokens

Corinna Hong · AI That Can Prove It’s Right: Verification as the Missing Layer in AI — Carina Hong · Feb 26, 2026 · at 28:35

Corinna Hong, founder and CEO of Axiom Math, discusses the scarcity of formal verification training data in Lean compared to general software code.

0:00 / 0:04exact quote · 4.2s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“It's only two-digit million number of tokens out there in the open, open world.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

As of February 2026, there are only tens of millions of Lean data tokens available in the open source world.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

The claim states that as of February 2026 there are only tens of millions ('two-digit million') of Lean data tokens available in the open source world. However, research data indicates that total open-source Lean tokens significantly exceed this figure. Mathlib4 alone contains between 138 million and 208 million tokens, and public GitHub Lean repositories add another 131 million tokens, bringing combined human-written Lean tokens to over 269 million to 339 million. Including public synthetic datasets like LeanNavigator (1 billion tokens) and ICML datasets (300 million tokens), total open-source Lean tokens reach hundreds of millions to over a billion.

Mathlib4 (LeanDojo Mathlib) extracted dataset contains 138 million to 208 million tokens.
Public open-source GitHub Lean repositories outside Mathlib contain an additional 131 million tokens, bringing combined human code to 269M-339M tokens.
Tens of millions (~57M) applied to older Lean 3 extractions from 2023, but is well below the actual count available in 2026.

Checked against

[1] arxiv.org [2] arxiv.org [3] emergentmind.com [4] openreview.net [5] emergentmind.com [6] github.com [7] arxiv.org

Lean 4 dataset tokens count mathlib "Mathlib" total number of tokens open source Lean Lean theorem prover dataset size tokens mathlib Lean Mathlib number of tokens dataset

assessed as of 2026-07-31

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

Lean

⚑ Spot an error? Report it

More from Corinna Hong

Assertion Not checkable as stated

AxiomProver is the first AI to solve research conjectures end-to-end

“It's probably the first AI to solve a research conjecture completely end-to-end and self-verifies. That means the output are fully verified, a hundred percent correct.”

Corinna Hong Feb 26, 2026 ▶ 4:56 AI That Can Prove It’s Right: Verification as the Missing Layer in AI — Carina Hong

Assertion Supported

AxiomProver achieved a perfect score on the 2025 Putnam math exam

“Eight within the time limit, and then 12 out of 12.”

Corinna Hong Feb 26, 2026 ▶ 4:02 AI That Can Prove It’s Right: Verification as the Missing Layer in AI — Carina Hong

Prediction Not checkable as stated

Today's AI can solve math problems that take human researchers months

“I think that we are at a threshold of mathematical renaissance, which is to realize that there are so many unsolved problems that will currently take, say, researchers months to crack, or even technical lemmas in those really longstanding conjectures that we b…”

Corinna Hong Feb 26, 2026 ▶ 6:14 AI That Can Prove It’s Right: Verification as the Missing Layer in AI — Carina Hong

Insight

True AI reasoning engines require verifiable rewards for intermediate proof steps

“If you want to have a reasoning engine that really truly masters at logic and mathematical reasoning, then you need to somehow get verifiable reward for the proof steps.”

Corinna Hong Feb 26, 2026 ▶ 19:52 AI That Can Prove It’s Right: Verification as the Missing Layer in AI — Carina Hong

Assertion Open · timeframe Feb 2029

Axiom's proof verifier is 100 times faster than open-source alternatives

“So a lot of the sort of like verify, verify proof is actually, you know, one of our prover tools that's about to be released, and that's actually a hundred times faster than The other counterparts that are the open source, like effort, cloud comparator.”

Corinna Hong Feb 26, 2026 ▶ 37:32 AI That Can Prove It’s Right: Verification as the Missing Layer in AI — Carina Hong

Assertion Not checkable as stated

AxiomProver autonomously proves theorems publishable in major mathematical journals

“Currently the batch of papers, Axiom Prover has autonomously proven and mathematicians have written You can probably get into Journal of Number Theory, Journal of Algebra, like that level.”

Corinna Hong Feb 26, 2026 ▶ 43:12 AI That Can Prove It’s Right: Verification as the Missing Layer in AI — Carina Hong

The open-source Lean dataset contains only tens of millions of tokens

More from Corinna Hong

Turn any episode into a week of clips.