NY
Closed
LDN
Closed
TKY
Closed
Your inflation
8 Oct

EquitiesMarket noteThursday 1 October 2026, 22:46

Gemini 4 Argon puts Google back in the AI race, but not in front

An early independent test ties Google's new model with OpenAI's best, behind Anthropic. Google may not need to lead. It sells computing power to its rivals, and owns part of one.

By The Notebook Desk

Alphabet began rolling out Gemini 4 Argon on Wednesday, its most advanced AI model yet, starting with a small group of cybersecurity partners, CNBC reported. Google says the model sets a new record in real-world software engineering and ties for first place on cybersecurity tests. Developers, businesses and consumers will get it later.

It is Google’s first new frontier model in more than seven months. Google has given no date for the wider launch, saying only “as soon as possible”. The first independent results put it back among the leaders. On the Intelligence Index of Artificial Analysis, which combines several benchmarks, Argon scored 53, The Decoder reported. That ties it with OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1. Anthropic’s newer Claude Opus 5.5 scored 58 and Claude Sonnet 5.5 scored 56.

Artificial Analysis Intelligence Index, score
Artificial Analysis Intelligence Index, score. Source: Artificial Analysis via The Decoder, 30 September 2026.
Claude Opus 5.558
Claude Sonnet 5.556
Gemini 4 Argon53
GPT-6 Astra53
Claude Fable 5.153
GPT-6.1 Sol52
Gemini 3.1 Pro30

Source: Artificial Analysis via The Decoder, 30 September 2026.The Macro Notebook

The jump from Google’s previous frontier model, Gemini 3.1 Pro, is 23 points. Argon also took first place in the Text Arena, where people rate answers from two models side by side. Its launch price is low for a model of this class: $2 per million input tokens and $10 per million output tokens, rising later to $4 and $20.

How Google fell behind

Nearly a year ago Gemini 3 put Google at the front of the race. This year it slipped. A planned Gemini 3.5 Pro was delayed, and in August Demis Hassabis, the co-founder of DeepMind, stepped back from running Google’s AI unit to become its chair, The Guardian reported. A former Google executive told the paper that Gemini had lost ground to models from Anthropic and OpenAI. Koray Kavukcuoglu, who is based in the US, now runs the unit.

Others left. Jeff Dean, Google’s former chief scientist and a co-lead of Gemini, left to start a new lab, Discovery Loop, with three other senior researchers, according to SemiAnalysis, a research firm. It wrote that Google’s odds of reaching the state of the art again “have dropped to zero”. Argon does not prove it wrong on leadership, but it does put Google back in the top group.

The race at the top is between OpenAI and Anthropic, as we wrote on Wednesday. Their annualised revenues are now measured in tens of billions of dollars. Google does not report Gemini revenue separately.

Google wins either way

Google may not need to lead to profit from the race. Its cloud business sells computing power to the companies that build AI, its rivals among them. Google Cloud revenue rose 82% to $24.8bn in the second quarter, the fastest growth of any part of Alphabet. In the same quarter Alphabet booked a net gain of $98.0bn, mostly unrealised, on its investments. Much of it came from its stakes in Anthropic and SpaceX, MediaPost reported.

The ties with Anthropic run both ways. In October 2025 Anthropic said it would use up to one million of Google’s TPU chips, in a deal worth tens of billions of dollars. When Anthropic’s Claude wins a customer from Gemini, Google can still earn from the chips Anthropic rents from it, and gain from the value of its stake.

Gemini already has a wide reach. The Gemini app had 950 million monthly users in the second quarter, and Google’s models processed 22 billion tokens a minute through its programming interface, Alphabet said. Nearly 90% of the Fortune 100 use Gemini Enterprise.

The market was calm. Alphabet’s shares rose 0.9% on Wednesday to $344.08 and fell 1.7% on Thursday to $338.24.

What to watch

The public launch of Argon is the first test: how fast it reaches paying customers, and at what price once the launch discount ends. The second is whether companies switch. A model that ties for third place on benchmarks wins business only if it is cheaper or easier to use than the leaders. The Decoder noted that Google’s apps still lag behind those of its rivals. The third is Alphabet’s results for the third quarter, which will show whether Cloud keeps growing at the same pace.

Disclosure

This is analysis and opinion, not investment advice.