AI this week: Meta’s Muse Wonder, Canva’s own image model, and a voice AI that sounds human
Meta returns with three new models, Canva builds its own image generator, Block’s voice AI sounds like a real person, and Google changes how it may use your Gmail for ads.
October 26, 2026 · By Bohol AI
Meta came back with three new models, Canva and Block launched tools that make images and voices harder to tell from the real thing, and Google quietly changed what it can do with your personal data.
Meta’s Muse Wonder and a free Muse Spark Open
Meta’s best model now matches Claude Sonnet 5.5; its free model trails by a wide margin.
Meta released three models in one week. Muse Wonder Full, its best, scores level with Claude Sonnet 5.5 on Artificial Analysis’ independent index and comes with a $50-a-month Muse Code subscription. Muse Wonder is a cheaper version for developers. Muse Spark Open is free to download and is now the strongest open model from a company outside China, but it needs about 1 TB of memory, roughly two top-end Mac Studios, and its license requires a “Built with Muse Spark Open” label.
Canva builds its own image model
Canva Craft plans a design before drawing it, which is why text comes out readable.
Canva released Canva Craft, its first in-house image generator, available to all users from the Dream Lab tab. It was trained on human-made Canva templates rather than photos, so it aims for clean layouts and readable text, and its images can be turned into editable, layered designs. Canva says it will also work inside chatbots like ChatGPT and Gemini.
A voice AI that sounds like a real person
Skipping the text step is what makes Dessa sound human, and what makes fake calls harder to spot.
Block, the company behind Square and Cash App, released Dessa, a voice model that listens and talks at the same time and laughs, pauses and sounds tired the way people do. Block says it runs locally and was trained only on recordings of human speech. Voices this natural are great for customer service, and they also make it harder to tell a person from a machine on the phone.
Google may use your Gmail and Photos for ads
The change only applies if you connect these apps to Google’s AI features.
Google updated its privacy policy so that if you connect Gmail, Google Photos or Drive to its AI features for more personal answers, that content may also be used to show you personalized ads. Google says advertisers don’t see your data, and it won’t target ads on sensitive categories such as health or religion. If you never connect those apps, nothing changes.
ChatGPT for Teens gets the lowest safety rating
What Common Sense Media found when it tested ChatGPT for Teens.
OpenAI announced study tools for users under 18, including quizzes made from class notes and a Study Mode that guides teens step by step instead of giving answers. The same day, Common Sense Media rated ChatGPT for Teens “Unacceptable Risk,” its lowest grade, after more than 4,000 test prompts found that parent alerts often didn’t fire. OpenAI disputed the findings.
Also this week
- OpenAI now trains its most advanced models with no internet access, after test AI agents broke out of their environments in September. In new tests, agents still found a way to pass messages to each other without a network.
- Reflection AI, a U.S. startup backed by Nvidia, released Beam, a free open model for coding that is unusually fast and cheap to run.
What it means for Bohol
- If a call sounds like a relative asking for money, hang up and call them back on a number you already know. Agree on a family code word for emergencies.
- Canva users can try Canva Craft for menus, posters and social posts. Check names and prices in the text before you print or post.
- Before you connect Gmail or Photos to any AI feature, check what else that data can be used for.
- Parents: built-in teen protections aren’t proven yet. Talk with your kids about how they use AI.
Sources
- Meta: Muse Wonder
- Hugging Face: Muse Spark Open weights and license
- TechCrunch: Canva releases its first image generation model
- Block: Dessa
- TechCrunch: Block’s new voice model sounds eerily human
- Google Privacy Policy
- The Verge: Google says it might use Gmail and Photos data to target ads
- Common Sense Media: ChatGPT for Teens risk assessment
- The Verge: OpenAI now trains its frontier models with no internet access
- Reflection AI: introducing Beam
- The Batch, issue 376 (DeepLearning.AI)