‘Ultra’ demos multi‑model flows

Dhaval Makwana’s post and short demo live on X under the handle @heyDhavall, where his threads and short-form posts are archived by thread-collector services. (threadreaderapp.com) Makwana’s public profiles list him as a UX/UI developer and content creator; his Linktree aggregates project links and his YouTube channel hosts how‑to and workflow videos. (linktr.ee) Anthropic’s Opus 4.6 — one of the models shown in the demo — was announced with a beta 1 million‑token context window and “adaptive thinking” controls intended for longer, multi‑step agent tasks. (anthropic.com) Google’s Gemini 3.1 Pro is presented on DeepMind/Google pages as a high‑capability, agentic model aimed at complex reasoning and interactive applications, with developer tooling like Google AI Studio and Antigravity cited on the product pages. (deepmind.google) Coverage of GPT‑5.4 frames it as the successor to GPT‑5.2 with expanded “thinking modes” and Pro variants that target deeper reasoning and agentic workflows, a set of features commentators say make it suitable as a specialist stage in multi‑model pipelines. (tweaktown.com) Public engineering examples already mirror the demo’s pattern: open GitHub workflow repos show stepwise routing where Claude/Opus handles architecture and logic, Gemini handles UI and frontend tasks, and a GPT variant is used for review and testing in a single pipeline. (github.com)