AI news6
Anthropic begins Nasdaq trading at US$111.50 a share
Anthropic began trading on Nasdaq under ANT on 19 November, opening at US$111.50 per share. Co-founders Dario and Daniela Amodei marked the listing at Nasdaq’s opening-bell ceremony.
nasdaq.com ↗Why it matters
The completed listing turns earlier IPO plans into public-market accountability for a major AI lab.
Gemini adds computer use on Apple Silicon Macs
Google began rolling out Gemini computer use on 19 November, letting it work across files, apps and websites in a separate background window. Access initially covers personal accounts aged 18-plus in U.S. English, on Google AI Plus, Pro or Ultra and macOS Sequoia 15 or later. Users choose app access, watch a live preview and can pause or approve key actions.
blog.google ↗Why it matters
This new Mac rollout makes local workflow automation available alongside Gemini’s earlier model releases, with access still limited by country and plan.
Canva adds two controls for generative tools and library content
Canva’s 19 November announcement adds a switch to hide generative AI features and a separate setting to show less AI-generated library content for Free and Pro users. The library filter reduces AI content in search and recommendations without changing existing designs. Background Remover and Magic Grab remain available when generative tools are switched off.
canva.com ↗Why it matters
The new controls give individual users a practical way to apply AI-use preferences during design work, while the library filter does not guarantee zero AI content.
Replit’s 32-agent coding test reports an 18.9-fold speedup
Replit’s 19 November study tested agent teams on performance-engineering workloads. On NumPy tasks, 32 Claude Opus 5.5 agents reached an 18.9-fold workload speedup in two hours, compared with 3.7-fold for one agent using less compute. GPT-6.2 Astra was the strongest solo model tested, but its results worsened as teams grew.
frontier.replit.com ↗Why it matters
This new benchmark shows that adding agents depends on task structure and coordination, rather than guaranteeing better results.
Cloudflare opens weights for 27B and 9B extraction models
Cloudflare released updated Clairic and Clairic-flash weights on 18 November, with models hosted on Workers AI and available through Hugging Face. The 27-billion-parameter and 9-billion-parameter models extract structured fields from text, images and PDFs, accepting up to four images or PDF pages per request. Cloudflare describes its accuracy results as internal, self-reported evaluations.
blog.cloudflare.com ↗Why it matters
The new open-weight release gives teams a hosting choice for document extraction, with independent accuracy checks still needed.
Vercel releases an experimental shared board for AI agents
Vercel launched Board on 18 November as an experimental message store and coordination tool accessed through a CLI or JavaScript SDK. Agents can claim tasks, post updates and keep results together in channels and threads. Vercel says its internal v0 software factory now merges more than half its pull requests using the system.
vercel.com ↗Why it matters
This new release makes Vercel’s internal coordination approach available to other teams, though its adoption figure is a company report.
AI in the nonprofit sector3
Mozilla’s Human-Scale AI call closes at 6:59pm AEDT today
Mozilla Foundation’s Human-Scale AI Incubator closes at 6:59pm AEDT on Saturday 21 November, equivalent to 20 November at 11:59pm Pacific Standard Time. Selected working-technology projects receive US$300,000 and 12 months of support from April 2027, with up to US$200,000 more subject to progress. Applicants need a charitable purpose and the ability to receive Mozilla grant funds, but need not be nonprofits.
mozillafoundation.org ↗Why it matters
Today is the final application window for this previously covered call, with follow-on funding conditional rather than guaranteed.
K-12 AI grants of up to US$600,000 close at 10:59pm AEDT
The K-12 AI Infrastructure Program’s October call closes at 10:59pm AEDT today, equivalent to 20 November at 11:59pm Anywhere on Earth. A US$1 million pool will fund up to two awards, capped at US$600,000 each, for open-source tools supporting AI-enabled formative assessment. U.S.-based nonprofits, universities and companies can apply, with outputs required to serve U.S. students.
k12-ai-infrastructure.org ↗Why it matters
This is the deadline-day reminder for the previously covered call, focused on giving teachers more visibility into and control over classroom AI.
South Africa’s research grant call includes AI competition topics
South Africa’s Competition Commission is accepting proposals for its 2027 Multidisciplinary Research Grant Programme until 27 November. Topics include competition in AI chips, cloud and data centres, and the AI value chain and adoption in South African markets. The call invites doctoral students, postdoctoral researchers and academics at South African tertiary institutions, plus experienced researchers in research institutions, but does not specify an award amount.
uct.ac.za ↗Why it matters
The live call offers an AI-policy research route for the academic and public-interest sector in South Africa.

