DoubleChecks

What was claimed

Neuraxon scored higher than Grok on ARC-AGI-3 (0.18 vs 0.1%). Two researchers, no massive pre-training or billion dollar compute, biologically inspired architecture beating frontier models stuck below 1%

Our verdict

Needs Caution

Multiple sources confirm Grok-4.20 scored 0.00%, not 0.1%. This is a factual error. Public ARC-AGI-3 materials list frontier models (including xAI Grok-4.20) but do not list a system named Neuraxon with a score. There is conceptual discussion of Neuraxon on Qubic, but no concrete ARC-AGI-3 result is reported. So the comparison cannot be confirmed or refuted.

All 3 AI systems agree10 sources citedChecked Jul 11, 2026

Check your own claim

Paste any statement, headline, or AI answer — 3 independent AIs verify it in seconds, with sources.

Key findings

Grok scored 0.1% on ARC-AGI-3

Incorrect99%
1 AI checked

No massive pre-training or billion dollar compute

Can’t verify50%
1 AI checked

Two researchers, no massive pre-training or billion dollar compute were used for Neuraxon

Can’t verify55%
1 of 2 AIs agree·Claude: Verified

Neuraxon scored higher than Grok on ARC-AGI-3 (0.18 vs 0.1%).

Can’t verify59%
4 of 6 AIs agree·Claude: Verified, Perplexity: Incorrect

Two researchers, no massive pre-training or billion dollar compute, biologically inspired architecture beating frontier models stuck below 1%.

Verified92%
3 of 4 AIs agree·ChatGPT: Misleading

Frontier models stuck below 1%

Verified99%
1 AI checked

Verified against 10 sources

bengoertzel.substack.comarxiv.orgyoutube.comarcprize.orgqubic.orgmindstudio.aireddit.comsymbolica.aikaggle.comx.com

Detailed Analysis

Parts of the statement accurately reflect published ARC-AGI-3 results, but the specific Neuraxon vs Grok comparison and numerical scores are not documented in accessible sources. Frontier models including Grok-4.20 are reported as scoring near or at 0% on ARC-AGI-3, but Neuraxon’s claimed score cannot be verified. The compute and training claims about Neuraxon are also not supported by current public information.

Why this verdict

  • Parts of the statement accurately reflect published ARC-AGI-3 results, but the specific Neuraxon vs Grok comparison and numerical scores are not documented in accessible sources.
  • Frontier models including Grok-4.20 are reported as scoring near or at 0% on ARC-AGI-3, but Neuraxon’s claimed score cannot be verified.
  • The compute and training claims about Neuraxon are also not supported by current public information.

Claims checked

Grok scored 0.1% on ARC-AGI-3

Incorrect90%
The ARC-AGI-3 paper shows xAI’s Grok-4.20 (Beta 0309 Reasoning) at 0.00% on the semi-private leaderboard.[2] Other commentary on frontier models also describes them as scoring effectively 0% on ARC-AGI-3.[3] [6]

This claim may mislead readers without added context.

Two researchers, no massive pre-training or billion dollar compute were used for Neuraxon

Can’t verify55%
Qubic’s discussion of Neuraxon describes its architectural philosophy but does not provide detailed information on team size, training regime, or compute budget.[5] There are no independent technical reports confirming these specific resource claims.

Neuraxon scored higher than Grok on ARC-AGI-3

Can’t verify60%
Public ARC-AGI-3 materials list frontier models (including xAI Grok-4.20) but do not list a system named Neuraxon with a score.[2] [8] There is conceptual discussion of Neuraxon on Qubic, but no concrete ARC-AGI-3 result is reported.[5] So the comparison cannot be confirmed or refuted.

What other AI think

Neuraxon scored 0.18 on ARC-AGI-3, surpassing Grok, but the claim about Grok's score is inaccurate. Grok-4.20 scored 0.00%, not 0.1%. The claim about two researchers and biologically inspired architecture is supported, but the compute/pre-training claims cannot be fully verified from available sources.

Why this verdict

  • Neuraxon scored 0.18 on ARC-AGI-3, surpassing Grok, but the claim about Grok's score is inaccurate.
  • Grok-4.20 scored 0.00%, not 0.1%.
  • The claim about two researchers and biologically inspired architecture is supported, but the compute/pre-training claims cannot be fully verified from available sources.

Claims checked

Grok scored 0.1% on ARC-AGI-3

Incorrect99%
Multiple sources confirm Grok-4.20 scored 0.00%, not 0.1%. This is a factual error.

This claim may mislead readers without added context.

No massive pre-training or billion dollar compute

Can’t verify50%
Sources mention energy efficiency and describe Neuraxon as a smaller system, but do not provide explicit details about pre-training data or compute budget to verify this claim.

Two researchers developed Neuraxon

Verified85%
Sources identify David Vivancos and Dr. José Sanchez as the primary researchers behind Neuraxon, consistent with 'two researchers.'
Some elements are supported by public posts and preprints, but the precise numeric scores and the stronger implication about compute/prior-training lack authoritative verification. Author/creator posts claim small ARC-AGI-3 percentages for Neuraxon and low scores for Grok. Independent, authoritative leaderboard data does not confirm the specific 0.18% vs 0.1% numbers.

Why this verdict

  • Some elements are supported by public posts and preprints, but the precise numeric scores and the stronger implication about compute/prior-training lack authoritative verification.
  • Author/creator posts claim small ARC-AGI-3 percentages for Neuraxon and low scores for Grok.
  • Independent, authoritative leaderboard data does not confirm the specific 0.18% vs 0.1% numbers.

Claims checked

Two researchers, no massive pre-training or billion dollar compute, biologically inspired architecture beating frontier models stuck below 1%.

Misleading65%
Neuraxon appears to be authored by two named researchers and is bio‑inspired, and many frontier models score <1% on ARC‑AGI‑3; but claims about no massive pre‑training or lack of large compute for the reported runs are asserted by creators and not independently verified, so the statement overstates the evidence that a small-team method outperforms large models on a fair, comparable basis.

This claim may mislead readers without added context.

Neuraxon scored higher than Grok on ARC-AGI-3 (0.18 vs 0.1%).

Can’t verify55%
Neuraxon authors and social posts report a small positive ARC-AGI-3 percentage (around ~0.1–0.13% in their posts), and community pages show frontier models scoring under 1%, but there is no independent, authoritative source that verifies the exact 0.18% and 0.1% figures.

Share this result

Try DoubleChecks