Summary
According to a post on X by @arcprize, ARC Prize reports that Google's Gemini 4 Carbon scored 65.1% on ARC-AGI-3 in its standard harness and 97.7% with Google's adapter, 18 points above its predecessor. The post also cites 49.6% on ARC-AGI-2 at $0.10 per task. The model is now generally available in GitHub Copilot, according to the same source. Separately, Google DeepMind and Artificial Analysis launched a new benchmark for agents doing visually complex work.
AI-assisted summary based on the listed source.
What happened
ARC Prize reports Carbon at 65.1% on ARC-AGI-3 in its standard harness and 97.7% with Google’s adapter, 18 points above its predecessor, plus 49.6% on ARC-AGI-2 at $0.10 per task. The model is now generally available in GitHub Copilot, and Google DeepMind and Artificial Analysis launched a new benchmark for agents...
Why it matters
Developers and teams comparing AI coding and agent tools may find these reported scores and the GitHub Copilot availability relevant to their evaluations. The post gives two ARC-AGI-3 figures, so readers should note which harness each one uses. Watch for independent results and for how the new visual agent benchmark performs.
Signal Intelligence
Signal Strength 91%
Technical label WATCH
Public Interest 57
Category RESEARCH
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 82
Practical Impact Score 26
Novelty Interest Score 80
Consequence Score 46
Curiosity Score 16
Shareability Score 65
Why this is here
VQV surfaced this signal because it is recent, relevant to Developer Tools, connected to Scobleizer News Pulse (X).
