FLUX 3 Video ships: 20-second clips with native audio, one API call
Black Forest Labs' first video model renders dialogue and SFX in one pass. Also: Qwen's 2.4T open weights drop this week; Sonnet 5's intro price ends Aug 31.
Copy markdown
20 seconds, sound baked in
FLUX 3 Video is Black Forest Labs' first video model, and it renders up to 20-second clips at 720p or 1080p/24fps with synchronized dialogue, sound effects, and ambience generated in the same pass — no audio editor, no post. It went live on the BFL API and dashboard on Aug 6.
Draft mode makes a weekend film pipeline cheap
On fal and Replicate you get five endpoints — text-to-video, image-to-video, first-last-frame, keyframes, and extend — each with a $0.06/sec draft mode to rough out shots before paying $0.17/sec HD or $0.29/sec Full HD (a 20s 1080p clip runs ~$5.80). Build this weekend: a prompt-to-storyboard-to-cut tool that never opens a video editor.
Qwen's first Max-class open weights land this week
Alibaba's Qwen3.8-Max — 2.4T params, 95B active, 1M context — is the first open-weight model in the Max tier, with weights promised on Hugging Face and ModelScope the week of Aug 10. It already matches GPT-5.6's $2/$6 per-Mtok API pricing; budget roughly 1.2TB for the 4-bit checkpoint before planning a local deploy.
Sonnet 5's intro price expires Aug 31
Claude Sonnet 5's promotional $2/$10 per-Mtok rate reverts to $3/$15 on Sep 1 — a flat 50% jump on both input and output. If Sonnet 5 is in your stack, you have about three weeks to audit spend or lean harder on prompt caching before the bill climbs.