Kandinsky 3
Apache 2.0 open text-to-image from Sber — strong on Russian cultural content.
About
Kandinsky 3 is Sberbank's open text-to-image line: 3.0 arrived November 2023 with Flan-UL2 text conditioning and a BigGAN-deep U-Net, and 3.1 followed in April 2024 as a distilled 4-step model with prompt beautification (an LLM that expands your prompt), a retrained inpainting model, IP-Adapter, and ControlNet support. Everything is Apache 2.0 — genuinely free for commercial use — with weights on Hugging Face under ai-forever. The training mix leans into Russian cultural content, which is its real niche versus Western models. Two honest caveats: global mindshare is small next to SDXL and FLUX, so community tooling is thinner, and the project has attracted political controversy in Russia, including a 2023 prosecutor inquiry over outputs and a 2026 vote against adding it to the domestic software registry.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: Kandinsky 2.1 reached FID 8.03 on COCO-30K (provider-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Kandinsky 3.0 (Nov 2023), 3.1 (Apr 2024, distilled 4-step) per Wikipedia; Sberbank ai-forever org.; Apache 2.0 — fully open for commercial use; weights on Hugging Face.; Niche strength: Russian cultural content in training mix.; Limitations: small global mindshare vs SDXL/FLUX; political controversies in Russia (2023 inquiry, 2026 registry vote)..