Kandinsky 4.0
Russia's open-source text-to-video model; 12-second HD clips from text or images.
About
Released December 2024, Kandinsky 4.0 marked Sberbank's Kandinsky line moving from images into video. Built on an MMDiT-style transformer with a 3D causal VAE (derived from CogVideoX), it generates 12-second HD video clips from text or image prompts, and Sber also shipped a video-to-audio companion model. In VBench evaluations and human preference tests it scored ahead of contemporaries like CogVideoX-1.5, Mochi 1.0, and Pyramid Flow at the time. The code and weights are open-source via the ai-forever GitHub organization.
Kandinsky is Russia's flagship generative visual model family, named after painter Wassily Kandinsky, and 4.0 was refined into Kandinsky 4.1 (a 14B DiT image model fine-tuned on artist-curated data, June 2025) and then the Kandinsky 5.0 family (November 2025) — the first built on flow matching, with 6B image, 2B video-lite, and 19B video-pro variants.
Limitations: as a late-2024 video model it lags current-generation video quality and prompt adherence; community tooling is thinner than for HunyuanVideo or Wan. Note the political context: Kandinsky 2.1 drew a Russian prosecutor inquiry over politically sensitive outputs, and Sber's models face Western sanctions constraints on compute — factor this into long-term support expectations.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
VBench: ahead of CogVideoX-1.5, Mochi v1.0, Pyramid Flow at release (vendor-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 64; Trending score 72/100 from researcher assessment.