DoubleChecks

What was claimed

Gemini 4 Argon scores 97.5% on ARC-AGI-3 making Google the second lab above 95%, major frontier breakthrough

Our verdict

Needs caution

This is an evaluative judgment rather than a directly measurable fact. The result is notable, but its significance is limited by the harness difference and incomplete Standard-harness evaluation.

Combined assessment13 sourcesChecked Oct 16, 2026

Why this verdict

  • This is a major frontier breakthrough.

    Can’t verify
    Two assessments reached this conclusion; one differed·Claude: Misleading

  • Google is the second lab above 95% on ARC-AGI-3.

    Verified
    Two assessments reached this conclusion; one differed·ChatGPT: Misleading

  • Gemini 4 Argon scores 97.5% on ARC-AGI-3.

    Verified
    Two assessments reached this conclusion; one differed·Perplexity: Misleading

Evidence reviewed

13 sources

Perplexity, Claude and ChatGPT sources are combined here.

x.comx.comarcprize.orgblog.googlesea.mashable.comdeepmind.googleblog.googleneuraltrust.aithedailystar.netapidog.com

Individual assessments

Assessment summary

  • The 97.5% figure is real only under ARC-AGI-3's Provider Adapter harness, not the Standard harness, where the score is 49.8% x.com.
  • The claim that Google is the second lab above 95% is supported by the ARC Prize listing, but the wording omits this important evaluation distinction x.com x.com.

Claims reviewed

Gemini 4 Argon scores 97.5% on ARC-AGI-3.

Misleading
Gemini 4 Argon scored 97.5% with the Provider Adapter harness. Its Standard harness score was 49.8%, so presenting 97.5% without naming the harness gives a misleading impression x.com.

This claim may mislead readers without added context.

This is a major frontier breakthrough.

Can’t verify
This is an evaluative judgment rather than a directly measurable fact. The result is notable, but its significance is limited by the harness difference and incomplete Standard-harness evaluation x.com.

Google is the second lab above 95% on ARC-AGI-3.

Verified
The ARC Prize results page states that Google DeepMind is the second frontier AI lab to score above 95% on ARC-AGI-3. This comparison refers to the Provider Adapter harness result x.com x.com.

Assessment summary

  • The factual claims about the 97.5% score and Google being the second lab above 95% are accurate.
  • However, the claim lacks critical context: the 97.5% score depends entirely on using the Provider Adapter harness, while the standard harness yields only 49.8%.
  • This substantial difference is material to evaluating whether this represents a "major frontier breakthrough" and makes the claim potentially misleading without this qualification.

Claims reviewed

Major frontier breakthrough

Misleading
The 97.5% score is contingent on the Provider Adapter harness. With the standard harness, performance drops to 49.8%, a 4.9x difference. This critical technical dependency is omitted from the claim, potentially overstating the achievement.

This claim may mislead readers without added context.

Gemini 4 Argon scores 97.5% on ARC-AGI-3

Verified
Verified by official ARC Prize results. The 97.5% score is achieved using the Provider Adapter harness.

Making Google the second lab above 95%

Verified
Confirmed by ARC Prize. Only OpenAI's GPT-6 Astra (99.9%) scores higher on ARC-AGI-3.

Assessment summary

  • The numerical score (97.5%) for Gemini 4 Argon on ARC-AGI-3 is reported by the ARC Prize results page but applies to a specific evaluation configuration (Provider Adapter harness).
  • Other public ARC-AGI-3 scores use a different, provider-neutral (Standard) harness where top models score much lower, so the claim omits an important context.
  • The qualitative claim of a "major frontier breakthrough" is subjective and unsupported by a single benchmark result alone.

Claims reviewed

This makes Google the second lab above 95%

Misleading
Other entrants (notably GPT‑6 Astra) have reported >95% on ARC‑AGI‑3 but typically under the same adapter-style harness; under the provider‑neutral Standard harness top models score far lower, so saying "second lab above 95%" is true under the adapter condition but omits that important harness-specific caveat arcprize.org arcprize.org.

This claim may mislead readers without added context.

This is a major frontier breakthrough

Can’t verify
Whether a single high harness-specific score constitutes a "major frontier breakthrough" is a value judgment and depends on broader replication, apples‑to‑apples comparisons, and community consensus not established by the cited result alone.

Gemini 4 Argon scores 97.5% on ARC-AGI-3

Verified
ARC Prize's published results list Gemini 4 Argon at 97.5% on ARC-AGI-3 under the Provider Adapter harness arcprize.org.

Check your own claim

Paste any statement, headline, or AI answer. Independent assessments check it and compare sources.

Share this result