Google DeepMind's newest model edges out Anthropic and OpenAI by a single point, though most people can't use it yet
Alphabet Inc. (GOOGL) price
Full chart and statsLoading chart…
Market cap $4.0T (shares as of 2026-09-30)
P/E (trailing) 17.0 (EPS to 2026-09-30)
Annual operating margin 32.0% (FY ended 2025-12-31)
Forward P/E Unavailable: no dated forecast source
Gloria daily close for session Oct 28, 2026. Market cap and trailing P/E use this close with the SEC share total and trailing diluted earnings, each dated next to its figure. Annual revenue, revenue growth, net income and margin show their own fiscal period and link the 10-K. Current context may postdate this article.
Alphabet Inc. and its peers
Token vs underlying stock
Current comparison for GOOGL. Historical observations may postdate this article.
GOOGLx token unit vs GOOGL share · USD
$325.97Oct 28 UTC
-10.2% over displayed period
● GOOGLx token┄ GOOGL stock: $324.89
Google DeepMind has a new model at the top of the leaderboard. Gemini 4 Carbon scored 59 on the Artificial Analysis Intelligence Index v5, enough to claim first place.
The model was unveiled on October 28, 2026. It hit that score at its high reasoning setting, and it arrived only weeks after its predecessor.
A one-point lead over the competition
The top of the index now looks like a photo finish. Gemini 4 Carbon sits at 59, Anthropic’s Claude Opus 5.5 trails at 58, and OpenAI’s GPT-6 Astra follows at 57.
The bigger story may be how far Google moved against itself. Gemini 4 Argon, released on September 30, 2026, scored 50 on the same index.
That makes Carbon a 9-point jump over its own sibling.
The Intelligence Index is a composite built from nine separate evaluations. Carbon leads in four of them, including HLE-Diamond and AutomationBench-AA.
Built with security teams in mind
Where Carbon really flexes is cybersecurity. The model posted 88.2% on SEC-bench Pro, which measures vulnerability discovery, and 71.0% on CWE-bench, which tests code patching.
It also scored 54.8% on ExploitGym.
Google’s research also points to gains in web development and secure code generation. The model reached 96.2% on GraphWalks, a benchmark tied to long-context comprehension.
There is one category where Carbon doesn’t take gold. On the separate Artificial Analysis Coding Agent Index, it scored 59 and ranks second, behind Claude Opus 5.5.
Access is limited, for now
Initial access goes to selected cybersecurity defenders through Google’s Fairwind Program. A public API release is still pending.
Standard pricing is set at $4 per million input tokens and $20 per million output tokens. Google is applying a 50% discount to that rate during the initial launch.
Background: a crowded release calendar
Carbon is the second Gemini 4 series model to be confirmed, following Argon’s September 30 debut.
What this means
For enterprise security teams, Carbon is the headline. A model that scores 88.2% on vulnerability discovery and 71.0% on patching is pitched directly at organizations drowning in unpatched code and short on staff.
Google’s decision to start with defenders, rather than the general public, is also telling. Models good at finding vulnerabilities could be just as useful to attackers, and gating access through Fairwind lets Google control who gets first crack.
Pricing is the other lever to watch. At $4/$20 per million tokens with a 50% launch discount, Google appears to be competing on cost as well as capability.
The key things to track next are the timing of the public API release and whether the launch discount holds once access widens.
Carbon’s coding agent result is worth keeping an eye on too. Agentic coding is where much of the enterprise spending is heading, and second place there leaves Google with a clear gap to close against Claude Opus 5.5.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

2 weeks ago
128








English (US) ·