Z-Image-Turbo

Pricing — 1 offering(s)

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

Capability profile

Operator guidance

Route here for cheap, fast generation at small model size — a strong default for high throughput. Step up to a larger model (e.g. HunyuanImage 3.0 or a FLUX pro tier) when you need maximum fidelity, complex composition, or finer detail.

Use cases

• High-volume, latency- and cost-sensitive generation
• Photorealistic images with Chinese/English text
• Open-weight self-hosting where a small, fast model is preferred

Limitations

• 6B distilled model — less headroom than frontier-scale models on complex scenes
• Capability ratings are qualitative. An independent benchmark score is now available: Artificial Analysis Text to Image Arena (artificialanalysis.ai/image/leaderboard/text-to-image, confirmed 2026-08-19, SCO-462 sweep) ranks 'Z-Image Turbo' #67 of ~147 models tracked, Elo 1129 — a distinct 'Z-Image Base' variant is separately listed at #103 (Elo 1064, not otherwise tracked in this registry). A different benchmark type than GenEval/T2I-CompBench, not a substitute for one
• Vendor/host descriptions are the main source for quality claims to date

Citations

• Z-Image-Turbo model card (Tongyi-MAI, Hugging Face)
• Z-Image-Turbo on SiliconFlow
• Z-Image Turbo — Artificial Analysis Text to Image Arena (independent Elo)