Google's Gemini 4 Argon hits 97.5% on ARC-AGI-3 but stumbles on standard tests
Source published ·Modelwire updated
Original coverage: The Decoder ↗·How Modelwire adds context
The development
Google's Gemini 4 Argon has reached 97.5 percent on ARC-AGI-3, positioning the company as only the second lab to breach the 95 percent threshold after OpenAI. However, the achievement masks significant fragility: performance collapses to 49.8 percent under standard testing conditions, revealing benchmark sensitivity rather than robust capability gains. Argon continues to underperform competitors in mathematical reasoning and code generation, suggesting Google remains behind on practical reasoning tasks despite headline-grabbing scores. The result underscores how benchmark selection shapes perceived progress in frontier AI development.
Modelwire’s AI-generated summary of coverage from The Decoder.
Modelwire analysis
Skeptical readOur AI-generated reading of the wider context and the next developments to watch.
The real story isn't the 97.5% result on ARC-AGI-3, it's that Google is selectively reporting benchmarks where Argon performs well while remaining silent about the domains where it still lags. The 49.8% collapse under standard testing reveals the benchmark itself, not the model, is doing the heavy lifting.
This contradicts the narrative from Google's October 13 launch coverage, which positioned Argon as a credible frontier competitor targeting reasoning-heavy workloads. The robotics positioning from October 15 (87% on ERQA) and the Chromium refactoring demo both relied on similar cherry-picked benchmarks and controlled environments. The pattern is consistent: Google showcases Argon in narrow, curated domains while avoiding head-to-head reasoning comparisons where Claude Opus 5.5 and Muse Wonder Full maintain leads.
If Google publishes Argon's performance on GPQA Diamond or the full ARC-AGI test suite without filtering, that signals honest capability assessment. If they don't release these numbers within 30 days, the 97.5% claim should be treated as marketing artifact, not evidence of frontier progress.
This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error
Coverage behind this analysis
These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.
·The Decoder
Google launches Gemini 4 Argon globally, trails top competitors in benchmarks
Google's Gemini 4 Argon enters the competitive frontier model race with aggressive pricing and broad distribution across API, consumer app, and enterprise channels. However, independent benchmarks place it behind Anthropic's Claude Opus 5.5 and Meta's Muse Wonder Full, signaling continued fragmentation in the top tier. The internal claim that Argon accelerates coding workflows from months…
MentionsGoogle DeepMind · Gemini 4 Argon · OpenAI · ARC-AGI-3
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. The Decoder originally reported this story as “Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months”. The full content lives on the-decoder.com. If you’re a publisher and want a different summarization policy for your work, see our takedown page.