CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins87% CONFIDENCEOverall confidence: 87%How well the pin's source and references back up its dates.Weighted average of how firmly 5 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
VentureBeat[4] reported on 8 December 2025 (5:03 pm PT) that Z.ai "has released its GLM-4.6V series" and OpenRouter[5] lists z-ai/glm-4.6v on 8 December; the Hugging Face[1] repositories were uploaded on 7 December UTC and docs.z.ai[2]'s launch case-study videos are labelled 1208.- Start
- Dec 8, 202590% CONFIDENCEfrom the source
GLM-4.6V and GLM-4.6V-Flash Released
Z.ai releases GLM-4.6V, a 106B vision-language mixture-of-experts model, and the 9B GLM-4.6V-Flash under the MIT license, the first GLM-V models with native function calling, trained for a 128K-token context and priced at $0.30 per million input tokens.
- Z.ai (Zhipu AI) released the GLM-4.6V series on 8 December 2025: GLM-4.6V, a 106B model (12B activated, per Z.ai's chart) for cloud and cluster use, and GLM-4.6V-Flash, a 9B model for local and low-latency use, both MIT-licensed and trained with a 128K-token context[1][2][4]
- The headline feature is native function calling in a vision-language model for the first time in the GLM-V line: images, screenshots and document pages can be passed straight into tools as parameters and tool outputs such as charts, search images or rendered pages are read back visually, closing the loop from perception to action[1][2]
- Z.ai reports state-of-the-art results among open models of comparable size on more than 20 multimodal benchmarks; in its chart GLM-4.6V scores 88.8 on MMBench V1.1, 85.2 on MathVista, 88.6 on Design2Code (GLM-4.5V 82.2) and 37.2 on OSWorld (35.8), while slipping on WebVoyager (81.0 against 84.4) and RefCOCO (88.6 against 91.3)[1][2]
- API pricing is $0.30 per million input tokens and $0.90 per million output tokens, GLM-4.6V-Flash is free and docs.z.ai also lists a FlashX tier; OpenRouter lists the same price with a 131,072-token context[4][5]
- Z.ai's model card admits that pure-text question answering still has room to improve, that the model can overthink or repeat itself on complex prompts, and that counting and identifying specific people remain weak[1]
Notable features
- Interleaved image-text generation that pulls in search and retrieval tools to fetch and curate extra pictures while writing, and visual web search with multi-step report generation[1][2]
- Frontend replication and visual editing: rebuilds pixel-accurate HTML, CSS and JavaScript from a UI screenshot and applies natural-language edits, including visual debugging[1][2]
- Long-context document understanding that Z.ai says covers about 150 pages of documents, 200 slides or a one-hour video in a single pass, plus multi-image reasoning and object detection returning labels and bounding boxes in JSON[2]
- The GLM-V family is described in the paper that introduced GLM-4.1V-Thinking and GLM-4.5V, whose later versions add the GLM-4.6V series with native tool use and a 128K window[3]
Benchmarks
A subset of rows from Z.ai's GLM-4.6V results chart (an image on the model card and docs.z.ai), thinking models; "/" not reported[1].
| Benchmark | GLM-4.6V (106B) | GLM-4.6V-Flash (9B) | GLM-4.5V (106B) | Qwen3-VL-8B-Thinking | Qwen3-VL-235B-A22B-Thinking |
|---|---|---|---|---|---|
| MMBench V1.1 | 88.8 | 86.9 | 88.2 | 84.3* | 90.6 |
| MMStar | 75.9 | 74.7 | 75.3 | 75.3 | 78.7 |
| MMMU (Val) | 76.0 | 71.1 | 75.4 | 74.1 | 80.6 |
| MMMU-Pro | 66.0 | 60.6 | 65.2 | 60.4 | 69.3 |
| MathVista | 85.2 | 82.7 | 84.6 | 81.4 | 85.8 |
| AI2D | 88.8 | 89.2 | 88.1 | 84.9 | 89.2 |
| Design2Code | 88.6 | 69.8 | 82.2 | 56.6* | 93.4 |
| OSWorld | 37.2 | 21.1 | 35.8 | 33.9 | 38.1 |
| AndroidWorld | 57.0 | 42.7 | 57.0 | 50.0 | / |
| WebVoyager | 81.0 | 71.8 | 84.4 | 47.7* | 30.4* |
| MMLongBench-Doc | 54.9 | 53.0 | 44.7 | 48.0 | 56.2 |
| OCRBench | 86.5 | 84.7 | 86.5 | 81.9 | 87.5 |
| ChartQAPro | 65.5 | 62.6 | 64.0 | 58.4* | 63.6* |
| RefCOCO-avg (val) | 88.6 | 85.6 | 91.3 | 89.3* | 92.4 |
| Ref-L4-test | 88.9 | 87.7 | 89.5 | 88.6* | 90.4 |
Buy onZ.ai