Qwen3.8 Max, Minimax H3, Shopify on Workers — AI Daily Aug 8

@jcartu reported Qwen3.8 Max is at parity with Kimi K3 in quality, slightly slower, with 1M context and vision at ~$40-60/mo or $250/quarter. Builders discussed multi-account rotations and anticipated GLM 5.2 and DeepSeek 4 Pro, while @weslindquist analyzed DeepSeek's price hike as pre-funding-round positioning. @jarvisballer learned GLM cloud needs 4-6 RTX 6000s to run properly.

@arielletolome benchmarked Minimax H3 at ~7 min per 15s clip on a single GPU (~34GB VRAM), with dual ComfyUI instances slowing to ~12 min from VRAM contention. Community compared 3090 (90s/5s), 5090, 4090, and RTX 6000 setups. Seedance 2.5 launched with 30s single-shot clips and 5K-char prompts at ~$10/video via Higgsfield. She also shipped new n8n useapi nodes.

@Anonymoushat is open-sourcing a Shopify clone on Rust CF Workers, Turso, Tinybird, and Durable Objects — bootstrapped from a compact BFF (Backend for Frontend) monorepo spec covering orders, customers, billing, and tax. Building it live on 150kb airplane wifi, targeting ~8 hours.

A paid 'call Jesus' offer prompted @arielletolome to spin up a Jesus voice agent same-day, while @filiuser proposed a Bible-grounded chat wrapper pulling scripture on the backend. @dragodimitrov noted EWTN's acquisition of Holy Habits validates the category. @justingacina and @rockdm also flagged Google's phone voice agent and Sesame as conversational AI benchmarks.

@tvrxm recommended Cloudflare's new transactional email service at $0.35/1K, surprising members who see CF becoming an AWS-style all-in-one stack (marketing emails not yet supported). @tounano detailed a Kanban-style harness where agents deterministically move jobs between lanes, contrasting with Fable-style dynamic graphs, and noted no off-the-shelf tool implements Stanford's STORM via harness orchestration.

Minimax H3 video gen speeds & GPU scaling

Qwen3.8 Max as a Kimi K3 alternative and model stacking

Paid Jesus-hotline offer and spiritual AI voice agents