Canva said on 29 October that its new model, Brush, is “up to 69% faster for general design editing tasks and 46% lower cost per edit task compared to Canva’s previous design agent”, according to its newsroom announcement.
Brush is a vision-language model built by Canva Research, the company’s in-house AI lab, and it now powers supported design edits in Canva AI for Canva Pro subscribers, Canva says. By our arithmetic, the 46% figure means an edit task that cost 100 units on the previous setup now costs 54.
Short explainer · Canva Brush
A vertical news Short narrated from the headlines and opening lines of this piece and four related pieces. It goes public at 12:00 UTC on 30 October.
Business Pill · Model Routing
An evergreen Business Engineer explainer on model routing and inference cost. It explains the general idea only and does not discuss Canva.
The key insight: Canva says Brush, a model it post-trained from an open-weights base, cuts the cost of an edit task by 46% against its previous design agent; by our arithmetic, Canva’s 400 designs a second is about 34.6 million designs a day.
What happened next (30 Oct): Figma posted an article on X at 19:51 UTC on 29 October, 34 minutes after Canva’s Brush post by our arithmetic, describing its own design-editing model. Figma says it trained one of its earlier Llama 3.3 70B models on more than 250,000 examples and got it to be competitive on editing tasks with top models at the time. Our follow-up, Figma Trained Llama on 250,000 Examples as Canva Ships Brush, has the details.
Canva describes Brush as a model that “sees your design, understands how it’s put together, and makes the changes you ask for, from a single line of copy to every slide in a deck”. It says Brush uses “an agentic loop that writes and runs code to adjust design elements” and then inspects the page to guide its next step.
The company says Brush “started from a leading open-weights model”, which it post-trained on real design and editing tasks and combined with “large-scale reinforcement learning”. Canva does not name the open-weights model in the announcement.
In Canva’s internal testing, the announcement says, Brush “matched or exceeded the quality of our previous setup on most editing tasks”, with particularly strong results in restyling designs and adapting them to different formats.
Cost per edit task, indexed to Canva’s previous design agent = 100: Canva Brush 54, after the 46% cut Canva reports. Canva also says Brush is up to 69% faster on general design editing. The index is our arithmetic. Source: Canva newsroom, 29 Oct 2026.
The Comparison Is With Canva’s Own Agent
Canva’s 69% and 46% figures compare Brush with “Canva’s previous design agent”. A post on X from TestingCatalog described the same figures as a comparison with frontier models; the Canva announcement we read does not say that. Canva does not say which models its previous design agent used, or give absolute costs per task.
Canva’s own post introducing Brush, above, had about 751,000 views when we read it on 30 October. It calls Brush “Our first vision-language model for design”.
The TestingCatalog post, above, summarises the launch and attributes the figures to Canva; its frontier-model wording is the part that differs from Canva’s announcement.
Why Cost per Edit Matters at Canva’s Scale
Canva ties the cost cut to volume. “With hundreds of millions of people creating billions of designs on Canva, even a few extra cents per task adds up quickly”, the announcement says. In its Craft announcement a week earlier, Canva said more than a quarter of a billion people use Canva “to create over 400 designs every second”. By our arithmetic, 400 designs a second is about 34.6 million designs a day; Canva does not say how many of those involve an AI edit.
Canva’s co-founder and COO Cliff Obrecht told TechCrunch in February that Canva reached $4 billion in annual recurring revenue by the end of 2025, with more than 265 million monthly active users and over 31 million paid users. TechCrunch reported that Canva was last valued at $42 billion in a share sale. Our Canva business model page covers how that revenue is made.
Two In-House Models in a Week
Brush follows Canva Craft, an image-generation model Canva announced on 22 October. Canva calls Craft “the first foundation model built and trained from scratch by Canva Research” and says it made Canva the world’s second most advanced AI lab for image generation as measured by DesignArena. The Brush announcement describes Canva Research as “our 160-person AI lab”.
As we read the announcement, Canva frames the model as a cost decision as much as a quality one: it says even a few extra cents per task adds up quickly at its scale.
As we read it, Canva now runs two in-house models announced a week apart, Craft for image generation and Brush for editing, while saying it will keep combining them with partner models.
By our arithmetic, the 46% cut means 54 units of cost where the previous agent used 100; Canva gives no absolute cost per task, so the dollar saving is not established.
Canva newsroom, Introducing Canva Brush, 29 October 2026
“With hundreds of millions of people creating billions of designs on Canva, even a few extra cents per task adds up quickly”
Three Implications
FOR CANVA PRO SUBSCRIBERS Canva says Brush now powers supported design edits in Canva AI for Pro plans, with wider availability coming soon.
FOR MODEL PROVIDERS Canva says it will bring together leading partner models with its own specialised models; it does not say how much work moves from partner models to Brush.
FOR INVESTORS TechCrunch reported Canva was last valued at $42 billion in a share sale; Canva gives no figure for what AI editing costs it today.
The framework puts it this way: “Close the SCARCE layer (keep proprietary) + Open the ABUNDANT layer (commoditize)”.
As we read it, Brush fits that pattern: Canva starts from an open-weights model and adds its own design data, reward system and professional designers’ judgement, per its announcement. The framework was published in March 2026 and does not discuss Canva.
What Is Not Established
The speed and cost figures are Canva’s own, measured against its previous design agent; Canva has not published how they were measured, the absolute cost per task or the base model. The “up to 69%” figure is a maximum, not an average. The 54-unit index and the 34.6 million designs a day are our arithmetic from Canva’s figures. Canva says Brush is available to Canva Pro subscribers, with wider availability coming.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Canva says Brush, a vision-language model it post-trained from an open-weights base, cuts the cost of an edit task by 46% and is up to 69% faster than its previous design agent. The comparison is with Canva’s own earlier setup, not with frontier models.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The Brush and Craft details and figures come from Canva’s newsroom announcements of 29 and 22 October 2026; revenue, user and valuation figures are from TechCrunch’s February 2026 interview with Cliff Obrecht. The 54-unit index and the 34.6 million designs a day are our arithmetic. We did not contact Canva.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
