So shocking, Gemini calls Opus, the era of super agents has truly arrived. From today on, Opus drives everything!
Yesterday, Musk actually announced that Grok Bot will outsource tasks to Opus 5.5.
Just one day later, at the Google Cloud launch event, Google directly integrated Claude Opus 5 and Sonnet 5.5 into the Gemini Business enterprise agent!
Antigravity CLI directly became Claude's agent framework, and Gemini was reduced to a supporting role.
The significance of this news may be greater than releasing ten Gemini 4s. Opus 5.5 has essentially become the configuration standard for super agent models.
From now on, we no longer need to find any relay stations, nor do we have to worry about account bans every day. As long as you connect to any mainstream super agent, you can directly call super intelligence!
This means that the tech giants who were fiercely competing over underlying large models in the previous generation have now all transformed into "super lobsters." The shell looks menacing, whether it's Google or Musk, but peel away this layer, and inside it's all Claude!
Next, let's see how domestic giants follow up.
All future approaches are essentially to make a beautiful shell and then connect to Opus 5.5. The so-called shouting "All in ASI" is actually "All in Opus 5.5."
Next, it depends on whose agent and whose workflow can get infinitely close to and squeeze out the capabilities of this strongest model!
The large model cold war has today turned into a battle for entry points.
Silicon Valley Brawl: A Joint Stranglehold on OpenAI?
No one expected that real business warfare would be like this.
The integration of Opus 5 and Sonnet 5.5 into the Gemini Business enterprise agent is also the first time in history that Claude models appear alongside Google's own offspring in the Gemini selector.
Musk and Google, two giants on the large model track, actually made tacit moves within 48 hours: not recognizing Musk's own foundation, not recognizing OpenAI's GPT, only recognizing Anthropic!
This sparked crazy ridicule from big Vs: OpenAI really has too few friends, everyone is teaming up with Anthropic.
Why are they willing to spend big money to buy each other's models and stuff them into their own high-end product lines?
The answer is simple: enterprise-level entry hegemony has overwhelmed the victory or defeat of the models themselves.
In this game of power, the underlying model can be outsourced, using whoever is stronger. However, the core workflow entry, user bills, and enterprise data permissions must be firmly controlled within their own systems!
The giants calculate very precisely: they would rather pay the inference fees to Anthropic than let customers flow to ChatGPT!
This is a joint stranglehold on OpenAI.
Google Launch Event Highlight: The Strongest Gemini is Actually Claude 5.5?
Let's focus on that dramatic "Gemini at Work 2026" launch event.
Google grandly launched the Gemini agent. Its concept is very advanced: all work starts from the current prompt window. It is embedded in Workspace, shuttling freely in Gmail, Docs, and code environments, possessing all your business context.
Google proposed six magnificent core architecture principles: unified agent, full-end access, cloud memory, multi-agent orchestration, deep context, and—model selection flexibility.
It is precisely this "model selection flexibility" that became the biggest joke of the event.
In the dropdown menu of the enterprise agent, the options listed simultaneously:
- Gemini 4 Argon
- Gemini 3.8 Flash
- Claude Opus 5
- Claude Sonnet 5.5
This is just too brilliant.
At the launch event, DeepMind executives crazily showed off how powerful their new tool Antigravity CLI combined with Gemini 4 Argon is. As a result, developers excitedly rushed into the system to check—sorry, Argon is marked "New," but you can't select it at all!
Pichai had previously revealed: Argon will first be used by Fairwind network defenders, and only after passing security checks will it be released.
Someone summarized it thus: Antigravity CLI has completely become a proxy framework for Claude, and Sonnet 5.5 equals the current strongest Gemini!
Developers helplessly complained: Thoroughly disappointed.
Thoroughly disappointed! The demo flooded the screen, but in agy you can't select Argon.
Who is the demo for? Keep whetting appetites? Why does Google fail to seize the opportunity every time? Why is it always so dilly-dallying?
In fact, Google's move seems to have lost face, but actually saved its core: everything is to retain customers.
In the past year, nearly 500 Google Cloud customers processed over a trillion tokens, and 90% of the Fortune 100 are using Gemini Enterprise. If they are not given the best model, they will buy OpenAI's enterprise version tomorrow.
It doesn't matter whose brain is used; the money and the gateway must belong to me. Once customers are cultivated, Gemini will naturally rake in money, sitting back and "collecting tickets."
Google stated bluntly:
Gemini is an agent, and which specific model it uses underneath is a separate choice.
Behind Musk's "bowing his head": the $84.5 billion computing power calculation
If Google's calculation is the enterprise gateway, then Musk's "adopting Claude" is a conspiracy of capital and computing power.
On October 7, Musk announced that Grok Bot will integrate Claude Opus 5.5, paired with multimodal Midjourney and Suno.
Many people mocked Musk, saying "Grok ultimately can't beat Claude, huge investment in vain," but the truth is chilling.
According to recently disclosed SpaceX financing documents, Anthropic is negotiating with SpaceX and is expected to purchase up to approximately $84.5 billion in computing power resources.
The false reputation at the model level is insignificant in the face of real money business.
On stage, Musk lavishes praise on Anthropic's Claude; behind the scenes, it's the massive returns from the supercomputing cluster.
At the same time, Grok users get the strongest reasoning capability on earth, firmly locking in platform stickiness.
Whether Google or Musk, they have seen through this underlying logic: large models are becoming infrastructure, and the intelligence lead period of a single model has shortened to months or even weeks.
Sacrificing ecosystem gateway and computing power commercialization to preserve the false reputation of model rankings is not the optimal choice.
But in all this prosperity, only OpenAI is absent.
The silent tacit understanding of giants: why deliberately sideline OpenAI?
When Musk and Google both implant Claude into their respective agent hubs, who is the most uncomfortable?
Undoubtedly, it's OpenAI.
The former "dragon-slaying youth" has now not only become the public enemy of Silicon Valley, but also fallen into the predicament of being isolated by major giants.
Looking around this big game, OpenAI is tasting the bitter fruit it planted.
First, the distrust brought by "full-stack dictatorship."
From GPT-4 to GPT-4o, to the o1 series and its own Operator agent attempts, Altman's ambition has never been to be a "behind-the-scenes model supplier," but to uproot the entire consumer and enterprise ecosystem.
OpenAI wants to do the base model, the C-end super app, the enterprise agent operating system, and even personally enter hardware and search.
This posture of ultimately treating all ecosystem partners as "potential prey" forces all the giants to be on guard.
By contrast, Anthropic is thriving everywhere, maintaining a relatively restrained positioning, more like an "arms supplier," and has naturally become the giants' preferred white glove for balancing OpenAI.
Second, there is the positional-war crisis in the enterprise market.
In the enterprise market, customers do not need a "super tech empire" that is moody, frequently adjusts pricing, and is always trying to claim the enterprise's core data for itself.
According to Reuters, Anthropic's prospectus states that over the next decade its infrastructure commitments with six partners will total at least $518 billion, of which about 80% is non-cancelable or unrelated to usage: Google at least $111.1 billion, Amazon $110 billion, Microsoft $31.4 billion, Broadcom-related equipment leases about $161.2 billion and largely non-cancelable, and AMD expected compute power exceeding $20 billion.
Anthropic's persona on AI safety, plus its compliance with cloud vendors' entry points, leaves it with almost no resistance on the B2B side.
The giants would rather work for their mortal enemies than hand a knife to OpenAI.
A silent encirclement is pushing OpenAI into becoming a "lone outsider."
What OpenAI now faces is a dimensional strike—
In terms of models, Opus 5.5 and Sonnet 5.5 can already fight GPT-6 to a draw. In terms of entry points, Google has Workspace, the world's largest office scenario, and Musk has X, the world's largest real-time information stream.
And OpenAI? It relies on neither side. Although its enterprise edition is strong, it is still an "add-on" tool, lacking a natural workflow foundation like Google's full suite.
Overseas analyst Lincoln pointed out incisively: it is now time for Altman to call Zuckerberg. OpenAI's only hope for a breakthrough is to lower its posture and quickly team up with Meta.
Meta has just launched its own enterprise platform and Business Agent, holds the massive traffic and business merchants of WhatsApp, Instagram, and Facebook, and has not yet taken sides on top-tier closed-source models.
If OpenAI cannot ally with Meta and deeply embed GPT's capabilities into Meta's social and commercial ecosystem, then it really may be completely sidelined in this "era of super agents" and reduced to an "API provider."
In this October, we witnessed history.
Large-model companies are losing direct control over end users, replaced by "super agent platforms."
The current main theme is already incomparably clear: All in on the agent entry point! Opus 5.5 takes over everything!
Will OpenAI be strangled to death by the entire industry, or can it join hands with Meta for a desperate counterattack?
Who can become China's version of the "super lobster"?
The good show has only just begun.
References:
https://x.com/Presidentlin/status/2108271837859791070
https://x.com/testingcatalog/status/2108270023319970086
https://x.com/ThomasOrTK/status/2108245829530325307
https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026
This article comes from the WeChat public account "Xin Zhi Yuan," author: ASI Revelation Record, editors: Aeneas David
