whatmodel.sh / image / Z-Image (Base)
Z-Image (Base)
↗ share
+ contribute
Z-Image (Base) is an open-weights image model from Alibaba (Tongyi-MAI), released January 2026. Its published size is 6B parameters. Licence: apache-2.0.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
The maker states 6B for the Z-Image family (S3-DiT), which lists Z-Image as one of the four variants; same figure as catalogued z-image-turbo (6). Covers the diffusion transformer per the family statement; text encoder not separately counted.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Z-Image (Base), out of 197 with a score, then more from Alibaba (Tongyi-MAI).
| # | model | score | where | ||
|---|---|---|---|---|---|
| 61 |  FLUX.2 [dev] | 58±1 | api $0.046/image +local | ||
| 62 |  Reve V1 | 58±4 | comparison only | ||
| 63 | Z-Image (Base) | 58±2 | api $0.010/image +local | ||
| 64 |  Wan 2.6 Image | 58±4 | api $0.030/image | ||
| 65 |  HunyuanImage 2.1 | 58±2 | api $0.10/image +local | ||
| more from Alibaba (Tongyi-MAI) | |||||
| 78 | Z-Image-Turbo | 55±1 | api $0.0050/image +local | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #47 of 126 · 98% of its score · 11 tests
Beats Wan 2.7 Image, Krea 2 Large, HiDream-O1-Image here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| QC-T2I-Bench (English, HCQ total) | 88.1% ✎ | ▄····▄█▄▄█▄█ beats 8/10 | 18% |
| UltraText Bench (Composite, bilingual dense text rendering) | 61.9% ✎ | ▁▂▂█▃▁▂▃·▁·▁ beats 14/22 | 16% |
| LongText-Bench (EN) lab | 93.5% ✎ | ▃·····▅██▅▃▃ beats 6/12 | 11% |
| LongText-Bench (CN) lab | 93.6% ✎ | ▃·······████ beats 7/12 | 8.3% |
| GenEval lab | 84.0% ✎ | ▇▃▅▇▁▄▅▃█▃▃▅ beats 32/39 | 7.8% |
| OneIG-Bench (CN) lab | 53.5% ✎ | ▃·▃█▃·▃··▃·▃ beats 7/8 | 6.9% |
| OneIG-Bench (EN) lab | 54.6% ✎ | ▃······▃▅█▅▃ beats 8/9 | 6.8% |
| CVTG-2K word accuracy lab | 86.7% ✎ | ▄▄▄█▄▄·▄···▄ beats 4/8 | 6.7% |
3 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| DPG-Bench lab | 88.1% ✎ | ▄·▂·▂▆█▄▂▄▂▂ beats 14/17 | 5.8% |
| TIIF-Bench testmini (long prompts) lab | 83.0% ✎ | ▄······█··▄▄ beats 1/4 | 5.5% |
| TIIF-Bench testmini (short prompts) lab | 80.2% ✎ | █···█···██·█ beats 1/4 | 5.5% |
$ where to run it
Through an API
api · fal.ai · $0.010/image · fal.ai ↗ · price sourceUSD for one 1-megapixel output: fal bills this endpoint $0.01 per megapixel (price = rate × 1 MP; fal's rounding for 1024×1024 ≈ 1.05 MP is not stated; larger outputs cost proportionally more). Product: fal.ai serverless text-to-image. API model: fal-ai/z-image/base (fal: 'the foundation model of the Z-Image family'; the Turbo endpoints are separate). Settings: 1 MP output, one image, no LoRA. Price basis: published per-megapixel list rate × 1 MP. Alibaba Model Studio sells only z-image-turbo. Currency: USD. Not an Alibaba direct tariff. fal.ai official model page, observed 2026-09-26.
Other ways to run it
version
- Run it as published · needs ~10 GB · from github.com/leejet
sd-cli --diffusion-model models/z-image-Q4_K_M.gguf --vae models/ae.safetensors --llm models/Qwen3-4B-Instruct-2507-Q4_K_M.gguf -p "A cinematic photograph of a solitary hooded figure walking through a rain-slicked metropolis at night, neon reflections on wet asphalt" --cfg-scale 5.0 -v --offload-to-cpu --diffusion-fa -H 1024 -W 512
Official weights and documentation: huggingface.co/Tongyi-MAI ↗
$ how the score was worked out · Tongyi-MAI (Alibaba) — Z-Imag… 64%, QC-T2I-Bench: Scalable Questi… 18%, UltraText Bench: A Comprehens… 16%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Tongyi-MAI (Alibaba) — Z-Image technical report · lab-reported | 9 | 64% |
| QC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing) | 1 | 18% |
| UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation (arXiv:2610.09823v1) | 1 | 16% |
| prior · pull toward the middle | 2% | |
| weighted average | = 58.0 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json