At its flagship “Canva Create” event, Canva announced its most significant platform overhaul since its 2013 launch. Dubbed Canva AI 2.0, the release marks a transition from simple AI-assisted design to a fully agentic and conversational system. By integrating new foundational models and workflow automation, Canva aims to become a central “operating system” for creative work.

The update follows Canva’s strategic acquisitions of Simtheory and Ortto earlier this month, which provided the underlying agentic orchestration and marketing automation technology.

The Four Pillars of Canva AI 2.0

Rearchitected from the ground up, the new platform is built around four core “intelligent” capabilities:

• Conversational Design: Instead of starting with a blank page, users describe a goal or structure via text or voice. Canva AI 2.0 generates a fully editable design with hierarchy, layout, and brand assets applied from the start.
• Agentic Orchestration: The system can now execute multi-step goals. For instance, a prompt like “Create a multi-channel campaign for our summer product launch” triggers the AI to coordinate across presentations, social posts, and email templates simultaneously.
• Object-Based Intelligence: Unlike traditional generative AI that outputs “flat” images, Canva’s new Design Model understands layers. Users can ask to “change just the headline font” or “swap the background image” without regenerating the entire design.
• Living Memory: A persistent context layer that learns a user’s style, tone, and brand preferences over time. It uses a custom “About Me” profile to ensure every output feels personalized to the specific team or creator.

New “Workforce” Workflows

Beyond design, Canva AI 2.0 introduces several agentic workflows designed to pull external data into the creative process:

Canva Code 2.0 & HTML Importing

In a major push toward interactive content, Canva Code 2.0 now supports HTML Importing. This allows developers and designers to describe an interactive idea (like a custom form or a web experience), which the AI builds into a responsive, cross-device artifact. Users can import AI-generated code from external tools and edit it directly within Canva’s visual interface.

Speed and Efficiency: The In-House Models

The platform is powered by a new suite of proprietary models developed by Canva Original Research and Exploration (CORE):

1. Canva Proteus: Optimized for style transfer; reportedly 2x faster than previous versions.
2. Canva Lucid Origin: A generative image model that is 30x cheaper and 5x faster than traditional frontier models.
3. Canva I2V: Specifically designed for cost-effective image-to-video generation.

“A decade ago, we moved design from the desktop to the browser. Today, we are moving from a tool you use, to a partner that works with you,” said Melanie Perkins, CEO of Canva. “Canva AI 2.0 is the first platform that doesn’t just help you design—it helps you do the work.”

Availability

Canva AI 2.0 is launching today as a Research Preview. It is initially available to the first one million users who sign up via the Canva homepage, with a wider global rollout planned for the coming weeks.

OpenAI has officially launched GPT-Rosalind, a specialized large-scale inference model designed to revolutionize life sciences research. Named after Rosalind Franklin, the British chemist whose work was central to the understanding of DNA structures, the model represents OpenAI’s deepest foray into specialized “reasoning” models for science to date.

The announcement sent ripples through the biotech sector, with specialized drug-discovery firms like Schrödinger Inc. and Recursion Pharmaceuticals seeing their stocks dip by more than 5% as investors weighed the impact of a general AI giant entering the niche field.

Core Capabilities: Beyond General Intelligence

Unlike standard language models, GPT-Rosalind is built on top of OpenAI’s latest internal “reasoning” architecture, specifically fine-tuned for biochemistry, genomics, and translational medicine.

• Hypothesis Generation: The model can synthesize evidence from millions of peer-reviewed papers to suggest novel drug targets, particularly addressing the “Phase II bottleneck” where many drugs fail.
• Experimental Planning: It can independently design multi-step laboratory protocols, query specialized biological databases, and suggest experimental pathways for gene therapy.
• Predictive Bio-Reasoning: In a collaboration with Dyno Therapeutics, GPT-Rosalind reportedly outperformed 95% of human experts in predicting the functions of unpublished RNA sequences.
• Codex Life Sciences Plugin: OpenAI is also launching a free research plugin for Codex, connecting scientists to over 50 external scientific tools and high-fidelity data sources.

Elite Partnerships and Enterprise Access

OpenAI is initially rolling out the model through its “Trusted Access” program to ensure the technology is used for legitimate human health improvements.

• Launch Partners: Major pharmaceutical giants including Amgen, Moderna, and Thermo Fisher Scientific have already integrated GPT-Rosalind into their R&D workflows.
• Non-Profit Access: The Allen Institute for Biomedical Research is among the first non-profit organizations to use the model for open-science initiatives.
• Compliance: The model features “high-precision flags” to monitor and block any potential misuse related to bioweapons or prohibited pathogens, a move coordinated with global biosafety agencies.

The Competitive Landscape: OpenAI vs. Google vs. Anthropic

The release of GPT-Rosalind is widely seen as a direct challenge to Google’s dominance in the space via AlphaFold.

Market Impact and Availability

While GPT-Rosalind is available as a Research Preview for select enterprise clients starting today, it marks a significant shift in OpenAI’s strategy. Following the launch of GPT-5.4-Cyber earlier this week, the company is clearly moving away from “one-size-fits-all” models in favor of vertical-specific agents that can handle high-stakes professional tasks.

“We are moving from an era where AI answers questions to an era where AI assists in the hardest parts of the scientific method,” said Joy Jiao, Head of Life Sciences at OpenAI. “Rosalind isn’t just a chatbot; it’s a collaborator that can plan the next six months of a lab’s work.”

The “first-mover” era of total dominance for OpenAI appears to be ending. According to the latest data from Apptopia and Similarweb, ChatGPT’s market share has seen a historic decline in the first quarter of 2026, falling below the 50% mark on mobile devices for the first time since its inception.

While ChatGPT remains the world’s most visited AI site, its once-impenetrable lead is being eroded by a combination of deep ecosystem integration from Google Gemini and the rising “agentic” capabilities of Claude and Perplexity.

The “22-Point Drop”: By the Numbers

The shift in the AI landscape over the last 12 months has been drastic. Data indicates that ChatGPT is no longer the default choice for a growing segment of the population.

• Web Traffic Decline: ChatGPT’s global web traffic share fell to approximately 60.4%–64.5% in early 2026, a significant drop from the 87% dominance it held in January 2025.
• Mobile App Plunge: The most striking data comes from U.S. mobile usage, where ChatGPT’s share of daily active users (DAU) cratered to 45.3% in January 2026, down from 69.1% a year prior.
• The Rise of Gemini: Google’s Gemini has been the primary beneficiary, with its market share surging from roughly 5% to over 21% in the same period, driven by its integration into Android and the Apple Intelligence partnership.

Why Is OpenAI Losing Ground?

Industry analysts point to three primary factors contributing to the decline:

1. Distribution Advantage: Competitors like Google and Microsoft (Copilot) have baked AI directly into the operating systems and productivity suites (Workspace, Office 365) that billions of people already use, removing the need for a standalone ChatGPT app.
2. Specialization: While ChatGPT remains a “generalist,” users are increasingly migrating to Perplexity for search-focused queries and Claude for high-nuance coding and creative writing, where Claude 4.7 is currently seen as the market leader.
3. Monetization & Burn: Reports indicate OpenAI is facing a projected $14 billion loss in 2026. To stanch the bleeding, the company has implemented stricter token limits and “Sora” delays, frustrating power users who are now exploring Anthropic’s more flexible “Mythos” and “Opus” models.

AI Chatbot Market Share Comparison (Q1 2026)

The “Apple Intelligence” Effect

The decline accelerated in early 2026 following the full rollout of Apple Intelligence. Although Apple initially partnered with OpenAI, the later integration of Gemini as an equal-tier option allowed millions of iPhone users to bypass ChatGPT entirely.

“We are witnessing the ‘commoditization’ of the LLM,” says a lead analyst at Seeking Alpha. “When the intelligence is built into your keyboard, your email, and your browser, the standalone brand matters less than the convenience of the ecosystem.”

Looking Ahead: The “Agent” Wars

OpenAI is expected to counter this trend with the upcoming launch of its “Operator” agentic system, aimed at reclaiming the lead in autonomous task completion. However, with Anthropic and Perplexity already shipping deeply integrated desktop agents for Mac and Windows this week, the window for OpenAI to regain its 80%+ market share appears to have closed permanently.

Perplexity AI has officially moved beyond the search bar with the launch of Personal Computer, a first-of-its-kind agentic AI assistant designed for the Mac. Announced at the company’s inaugural developer conference, the system transforms a user’s Mac into a persistent “digital worker” capable of executing complex workflows, interacting with local files, and managing native applications 24/7.

The rollout marks Perplexity’s transition from an AI-powered search engine to an “AI operating system,” where the machine focuses on user goals rather than just processing isolated commands.

Key Features of the ‘Personal Computer’ Agent

Unlike standard AI chatbots, Personal Computer operates at the interface layer, meaning it can “see” and “interact” with your screen much like a human assistant.

• Local-Cloud Hybrid Architecture: While high-intensity reasoning happens in the cloud, the agent operates locally on your Mac. It can access iMessage, Apple Mail, Notes, and local file directories to find information or execute tasks.
• Persistent Execution: Because it is designed to run on dedicated hardware (optimally a Mac mini), the agent remains active even when you step away. It can continue researching a topic, updating a database, or monitoring a news feed in the background.
• Command with Shortcuts: Users can trigger the agent using a double-tap of the CMD keys. For example, it can “read” your current to-do list in the Notes app and begin checking off items by opening relevant apps.
• Security & Control: Every autonomous action requires a user confirmation via an integrated audit trail. It also features a “kill switch” for immediate termination of all active processes and 2FA via iPhone for remote management.

The “Personal” Cost: Pricing and Tiers

Perplexity is positioning this as a high-end tool for power users and enterprise professionals, resulting in a new premium tiering structure.

The Developer Play: New APIs

Alongside the Mac launch, Perplexity released four foundational APIs that power the Personal Computer experience for third-party developers:

1. Search API: For real-time, cited web data.
2. Agent API: A managed runtime for agentic workflows.
3. Embeddings API: For local file retrieval.
4. Sandbox API: A secure environment for code execution.

Why It Matters

The launch is a direct shot at the “agentic” ambitions of competitors like Microsoft’s Copilot and Apple Intelligence. By leveraging the visual interface of macOS rather than relying solely on APIs, Perplexity’s agent can theoretically work with any application, including legacy software or specialized tools that lack modern integrations.

“A traditional operating system processes commands; an AI operating system focuses on goals,” said Aravind Srinivas, CEO of Perplexity. “Personal Computer is the first step toward a world where your computer works for you even while you sleep.”

Anthropic has officially launched Claude Opus 4.7, its most sophisticated and capable AI model to date. Released yesterday, the new model is designed specifically for high-stakes enterprise workflows, advanced software engineering, and complex, multi-step agentic tasks.

While it is the current public flagship, Anthropic notably confirmed that Opus 4.7 is “less capable” than Claude Mythos, a restricted model the company deemed too advanced for general public release due to cybersecurity and safety considerations.

Key Advancements in Opus 4.7

Opus 4.7 introduces a “hybrid reasoning” architecture that allows it to self-verify its work before delivering a final response.

• Autonomous Software Engineering: In internal tests, Opus 4.7 built a complete Rust-based text-to-speech engine from scratch, including neural models and SIMD kernels, and verified its own output against a reference.
• High-Resolution Multimodal Vision: The model now supports images up to 2,576 pixels on the long edge (~3.75 megapixels), a 3x improvement over prior versions. This enables pixel-perfect analysis of dense screenshots, complex UI diagrams, and technical blueprints.
• Adaptive Thinking & “Extra High” Effort: A new “xhigh” (extra high) effort setting allows users to trade latency for deeper reasoning on “impossible” problems.
• 1-Million Token Context: The model maintains high performance and instruction-following across its full 1M token window, making it ideal for analyzing massive codebases or multi-day legal projects.

Performance Benchmarks

Opus 4.7 has reclaimed the top spot on several key leaderboards, particularly in the “humanity” and coding categories:

Pricing and Tokenizer “Sticker Shock”

While Anthropic has kept the official sticker price identical to Opus 4.6, developers are noticing a change in actual costs due to a new tokenizer.

• List Price: $5 per 1M input tokens / $25 per 1M output tokens.
• The Token Inflation: The new tokenizer produces 1.0x to 1.35x more tokens for the same amount of text. This means that while the rate is the same, the bill for a specific prompt could be up to 35% higher.
• Savings: Anthropic continues to offer up to 90% discounts for prompt caching and 50% for batch processing.

Availability

Claude Opus 4.7 is available starting today via:

• Claude.ai: For Pro, Max, and Team subscribers.
• API Platforms: Native Anthropic Console, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
• Claude Code: The CLI tool has been updated to default to the “xhigh” effort level for all premium users.

Safety Note: Despite its power, Anthropic reports that Opus 4.7 is “more reliably honest” than its predecessors, with a significant reduction in hallucination rates and a stronger resistance to “jailbreak” or prompt injection attacks.