whatmodel.sh / image / GLM-Image
GLM-Image
↗ share
+ contribute
GLM-Image is an open-weights image model from Z.ai, released January 2026. Its published size is 16B parameters. Licence: mit.
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of GLM-Image, out of 197 with a score, then more from Z.ai.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 100 |  Recraft V3 | 50±2 | api $0.040/image | ||
| 101 |  FLUX.1 Kontext [pro] | 50±2 | api $0.040/image | ||
| 102 | GLM-Image | 50±2 | api $0.015/image +local | ||
| 103 |  Step1X-Edit v1.2 Preview | 50±5 | comparison only | ||
| 104 |  Runway Gen-4 Image | 50±3 | api $0.050/image | ||
| more from Z.ai | |||||
| 142 | CogView4-6B | 41±6 | comparison only | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #70 of 126 · 90% of its score · 9 tests
Beats Recraft V3 here, though they rank higher overall. Trails Runway Gen-4 Image here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena text-to-image | 1012.1 ✎ | ▂▂▃▄▇█▄▂▄▂·▂ beats 6/75 | 58% |
| QC-T2I-Bench (English, HCQ total) | 83.9% ✎ | ▄····▄█▄▄█▄█ beats 2/10 | 25% |
| Qwen-Image-Bench (Q-Judger, v1) | 48.2% ✎ | █▅▅·▅▅█·█··▃ beats 0/17 | 6.3% |
| DPG-Bench lab | 84.8% ✎ | ▄·▂·▂▆█▄▂▄▂▂ beats 4/17 | — |
| LongText-Bench (EN) lab | 95.2% ✎ | ▃·····▅██▅▃▃ beats 9/12 | — |
| LongText-Bench (CN) lab | 97.9% ✎ | ▃·······████ beats 12/12 | — |
| OneIG-Bench (EN) lab | 52.8% ✎ | ▃······▃▅█▅▃ beats 3/9 | — |
| OneIG-Bench (CN) lab | 51.1% ✎ | ▃·▃█▃·▃··▃·▃ beats 5/8 | — |
1 more test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CVTG-2K word accuracy lab | 91.2% ✎ | ▄▄▄█▄▄·▄···▄ beats 8/8 | — |
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #43 of 62 * · 8.8% of its score · 1 test
Beats GPT Image 1 Mini, FLUX.1 Kontext [max], Qwen-Image here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GEditBench v2 (overall Elo) | 979.0 ✎ | ▃▂····▃█▂▃▂▃ beats 3/14 | 8.8% |
$ where to run it
Through an API
api · $0.015/image · docs.z.ai ↗ · price sourceUSD per generated image on the Z.ai API (all Z.ai list prices in USD). Product: Z.ai image generation API. API model: glm-image. Settings: one image; recommended sizes such as 1280×1280. Price basis: published list rate. Currency: USD. Z.ai official pricing page, observed 2026-09-25.
Other ways to run it
version
- Run it as published · needs ~38.9 GB, NVIDIA only · from github.com/zai-org
uv run --with git+https://github.com/huggingface/diffusers --with git+https://github.com/huggingface/transformers --with accelerate --with torch python -c "import torch; from diffusers.pipelines.glm_image import GlmImagePipeline; p=GlmImagePipeline.from_pretrained('zai-org/GLM-Image', torch_dtype=torch.bfloat16, device_map='cuda'); p(prompt='a cat holding a sign', height=1024, width=1024, num_inference_steps=50, guidance_scale=1.5).images[0].save('out.png')"
Weights and documentation: huggingface.co/zai-org ↗
$ how the score was worked out · LMArena leaderboard dataset 58%, QC-T2I-Bench: Scalable Questi… 25%, GEditBench v2: A Human-Aligne… 9% and 1 more
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 |
|---|---|---|
 LMArena leaderboard dataset | 1 | 58% |
 QC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing) | 1 | 25% |
| GEditBench v2: A Human-Aligned Benchmark for General Image Editing | 1 | 9% |
| Qwen-Image-Bench: From Generation to Creation in Text-to-Image Evaluation | 1 | 6% |
| prior · pull toward the middle | 2% | |
| weighted average | = 50.0 |
not counted (6)
| zai-org/GLM-Image — committed benchmark documentation | CVTG-2K word accuracy [correct] | 91.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| zai-org/GLM-Image — committed benchmark documentation | LongText-Bench (EN) [correct] | 95.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| zai-org/GLM-Image — committed benchmark documentation | LongText-Bench (CN) [correct] | 97.9% | fewer than three models have this test yet, or this source ran it for this model alone |
| zai-org/GLM-Image — committed benchmark documentation | OneIG-Bench (EN) [correct] | 52.8% | fewer than three models have this test yet, or this source ran it for this model alone |
| zai-org/GLM-Image — committed benchmark documentation | OneIG-Bench (CN) [correct] | 51.1% | fewer than three models have this test yet, or this source ran it for this model alone |
| zai-org/GLM-Image — committed benchmark documentation | DPG-Bench [correct] | 84.8% | fewer than three models have this test yet, or this source ran it for this model alone |
$ published speed measurements · 3
Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.
| NVIDIA H100 | 252.59 s/image | diffusers · 2048×2048; text-to-image; batch size 1; fixed prompt ~1000 tokens Exact end-to-end timing. Dtype and H100 memory capacity not specified by this table. | source ↗ | |
| NVIDIA H100 | 64.34 s/image | diffusers · 1024×1024; text-to-image; batch size 1; fixed prompt ~1000 tokens Exact end-to-end timing. Dtype and H100 memory capacity not specified by this table. | source ↗ | |
| NVIDIA H100 | 27.3 s/image | diffusers · 512×512; text-to-image; batch size 1; fixed prompt ~1000 tokens Exact end-to-end timing. Dtype and H100 memory capacity not specified by this table. | source ↗ |
$ 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