Gemini 3 Arrives: Google's Most Capable Model Tops Every Major Benchmark
Announced at Google Cloud Next 2026 in Las Vegas on 22 April, Gemini 3 Pro Preview has taken the top position on the LMArena Leaderboard with a score of 1,501 Elo, ahead of all current competitors by a meaningful margin. It scores 91.9% on GPQA Diamond and 37.5% on Humanity's Last Exam without tools, placing it squarely in PhD-level reasoning territory. The model is live now as gemini-3-pro-preview on the Gemini API and in Google AI Studio, with Gemini 3.1 Pro also available in preview as the most advanced reasoning variant.
For production workloads where latency matters, Gemini 3 Flash is the practical option: it delivers a 15% accuracy improvement over Gemini 2.5 Flash and is optimised specifically for high-frequency agentic workflows. If you are currently evaluating frontier models for coding, research or complex reasoning tasks, the benchmark gap between Gemini 3 and the current field is wide enough to warrant a direct test.
Gemini Enterprise Agent Platform: Google's End-to-End Control Plane for Agentic Workflows
Announced from the Cloud Next keynote stage, the Gemini Enterprise Agent Platform is Google's most ambitious enterprise product release to date. It brings together four components: an Agent Designer for authoring agents visually without writing orchestration code; an Inbox for tracking long-running agent tasks; a Skills system for building composable, reusable capabilities; and a Knowledge Catalog that continuously enriches enterprise data by profiling logs and business context in the background.
Beneath it sits the Agentic Data Cloud, which connects a cross-cloud Lakehouse with the Knowledge Catalog to give agents grounded access to live business data rather than static snapshots. For teams building agentic workflows on Google Cloud, this is the unified control layer Google has been building towards: a single place that governs what agents can do, what data they can access, and how outputs are verified before surfacing to users.
Google Workspace Gets AI-Native Intelligence and Chrome Gains Auto-Browse for Enterprise
For anyone who spends most of their working day inside Google's productivity suite, the Workspace Intelligence announcement at Cloud Next 2026 is worth paying attention to. The system draws on Gmail, Calendar, Chat and Drive data to automate routine tasks, suggest follow-ups and surface relevant context without requiring manual search. It is presented as a production feature rather than a research preview, meaning enterprise admins can begin rolling it out to teams immediately.
In the same announcement window, Google confirmed that Gemini-powered auto-browse is coming to Chrome for enterprise users. Workers can delegate repetitive browser tasks (research, form filling, data entry) to an AI agent operating directly within Chrome, without leaving the browser or setting up a separate agent environment. Taken together, these two features represent a meaningful shift in how Google positions Workspace: less a set of applications, more an AI-assisted operating layer for knowledge work.
Google Confirms Gemini Will Power the Next Generation of Apple Siri at WWDC 2026
Google Cloud CEO Thomas Kurian confirmed from the Cloud Next 2026 keynote stage that Apple has selected Google as its preferred cloud partner for the next generation of Apple Intelligence features. Gemini technology will underpin a more personalised Siri, expected to be unveiled at WWDC 2026 on 8 June. The partnership had been hinted at in January 2026, but this was the first time either company confirmed it publicly from a major stage.
For anyone tracking the competitive landscape between AI assistants, the practical implication is significant: Siri's reasoning backbone is switching from Apple's internal models to Gemini, bringing it closer in capability to GPT-powered assistants whilst consolidating Gemini's reach across an enormous installed base of iPhone users. The announcement surfaced via the Cloud Next livestream before mainstream press picked it up.
Google announced a $750 million fund to help its 120,000-member partner ecosystem sell AI agents to enterprise clients. Eighth-generation TPUs were also confirmed, alongside the Virgo Network, a new megascale data centre fabric underpinning Google's AI Hypercomputer infrastructure. Google Cloud Press Corner, 22 Apr
ChatGPT Images 2.0 Rolls Out to All Users with O-Series Reasoning and 2K Resolution
gpt-image-2 is now rolling out to all ChatGPT users from 22 April, and the headline detail is what makes it structurally different from previous image models: it is the first to incorporate O-series reasoning capabilities. Before generating anything, the model researches layout, plans structure and verifies output coherence, producing markedly better results on complex compositions, multilingual text rendering, infographics and multi-image batches. It supports 2K resolution and can generate up to eight coherent images from a single prompt whilst maintaining character and object continuity across the batch.
Thinking mode, which unlocks web search and layout reasoning, is restricted to paid Plus, Pro, Business and Enterprise tiers. Developer API access is expected in early May. Within twelve hours of launch the model had claimed the top spot across every category on the Image Arena leaderboard by a 242-point margin. If you build marketing or product workflows that depend on image generation, this capability warrants evaluation now rather than waiting for the API.
Codex Labs Launches: OpenAI Places Its Own Experts Inside Enterprise Organisations
Announced on 21 April, Codex Labs is OpenAI's new service for getting enterprise teams from early experimentation to repeatable production deployment of Codex. Rather than selling a product and leaving integration to the customer, OpenAI is placing its own experts directly inside organisations through hands-on workshops and working sessions. Consulting partners include Accenture, Capgemini, Cognizant, Infosys, PwC and Tata Consultancy Services.
The timing provides useful context: OpenAI reports that Codex is now used by more than four million developers per week, up from three million just two weeks earlier. That growth rate suggests the bottleneck is no longer awareness but implementation complexity. For teams that have been evaluating Codex but stalled on the integration side, Codex Labs is the supported on-ramp that previously did not exist.
Industry Themes
The dominant story today is Google's wholesale announcement of a full agentic stack at Cloud Next 2026. In a single day, Google released a frontier model (Gemini 3), an enterprise agent platform, an AI-native workspace suite and confirmed a partnership that will put Gemini inside Apple's Siri. Taken together, these announcements suggest Google is no longer playing catch-up to OpenAI; it is attempting to define the entire vertical layer of AI tooling for enterprise, from the underlying model through to the orchestration, data and interface layers. The breadth of the Cloud Next announcements is the clearest signal yet that Google's AI strategy is vertically integrated rather than product-by-product.
The second theme is reasoning applied to image generation. OpenAI's gpt-image-2 brings deliberate planning to visual output for the first time, letting the model think about layout, structure and consistency before generating a single pixel. This mirrors the shift that happened in text models when O-series reasoning was introduced, and signals that the gap between "generate something that looks right" and "generate something structurally correct" is finally closing. Topping the Image Arena leaderboard by 242 points within twelve hours of launch is a hard number rather than a marketing claim.
Third, the Gemini/Siri confirmation from Google's own keynote stage is a structural shift for the consumer AI assistant market. The two largest consumer AI ecosystems outside OpenAI are now formally collaborating rather than competing, which consolidates Gemini's reach to an enormous iPhone installed base. It also raises a pointed question for Microsoft: Copilot is increasingly caught between Gemini expanding into Apple's ecosystem and ChatGPT deepening its grip on Microsoft's own productivity suite.