Jev AI by TypeSafe: System One model for typed calibrated decisions—Choice/Score/Noul, $0.042/MTok input, free outputs. Docs-backed review.

Kling 4.0: Kuaishou Kling AI video model launched Oct 15, 2026—30s native clips, 10 keyframes, Omni Reference (15 assets). Specs from official docs.

FLUX 3 Image Klein: Black Forest Labs 5.6B open weights for T2I, editing, 10 refs, native 4K, ComfyUI day-one, Kommunity License under $5M revenue.

ComfyUI MOSS-Image: OpenMOSS 5B Apache-2.0 T2I with bilingual text, layout boxes, SFT/Base checkpoints, and official templates (0.41.3+).

Filee: free MIT offline file converter for Win/macOS/Linux—drag-drop donut UX, 240+ formats, HWP/Office without upload. KnifeLemon/Filee on GitHub.

Concat video editor: open-source local CapCut alternative (AGPL) with Whisper captions, TTS, cutout, 4K export on Win/macOS/Linux/Android. Not OpenCut.

OmniRoute: MIT open-source AI gateway with one OpenAI-compatible endpoint, multi-provider failover, free-tier dashboard, Docker/npm. Local-first.

Seedance 2.5: ByteDance AI video model for up to 30s synced audio-video, 50 multimodal refs, timestamp editing. Access via Jimeng, Dreamina, ModelArk.

Datalab Indra OCR: ~1M-param model, ~49 pages/sec on 32 CPU cores, markdown/JSON with layout boxes. Install, outputs, limits vs Tesseract/Paddle.