June 23, 2026 — curated links and takeaways.
1. Seedance 2.5: ByteDance's 30-Second Native 4K AI Video Model
ByteDance unveiled Seedance 2.5 at Volcano Engine Conference 2026 with native 4K, 30-second single-pass clips, 50 reference inputs, and 20% better prompt accuracy; launching early July. Creators and studios face a significant capability jump—native 30-second generation eliminates stitching friction, directly competing with Kling 3.0 and Runway on speed and quality.
2. Perplexity AI Model Council Feature Expands Multi-Model Comparison
Perplexity launched 'Model Council' on February 5, 2026, letting users compare outputs from GPT-5.2 and Claude 4.6 side-by-side; now expanding free Pro access to students. This is a direct challenge to Google and Bing's search dominance—practitioners building on single-model backends now see consumer demand for model choice built into search UIs.
3. Browser Automation and AI Scraping Tools Mature for Production Workflows
Selector Forge, Tabstack (Mozilla), and Firecrawl are becoming standard for QA and structured data extraction on dynamic sites, with MCP server support and multi-step automation. Builders integrating AI agents into ops workflows now have mature open tooling for reliable web interaction—reducing API dependency and improving cost per automation run.
4. AI Agents Running on Cloud Infrastructure Without Local File Access
Happycapy demonstrates agent-like apps built on cloud data and APIs without local file access; reviewers note sandbox isolation, built-in analytics, and secure in-browser UI. Teams building internal AI agents should evaluate cloud-first architectures over local file systems—reduces deployment friction but locks state to host platform.