← All models

Glyph

Published Publisher Zhipu AI

Overview

Glyph is a research framework and vision-language model for long-context understanding. It renders long text as images and processes those images with a vision-language model, aiming to reduce computation and memory use while retaining semantic information. The published model is built on GLM-4.1V-9B-Base. (huggingface.co)

Capabilities and uses

Glyph can support long-document understanding, code analysis and other long-context reasoning tasks. In the authors’ experiments, it achieved approximately 3–4× input-token compression and roughly 4× faster prefilling and decoding on the tested workloads; these are experimental results, not guaranteed performance. (arxiv.org)

Limitations

Performance can vary with rendering settings such as resolution, font and spacing, and may generalize poorly to substantially different rendering styles. Training principally targeted long-context understanding; performance on broader tasks remains to be studied. (huggingface.co)

Availability and date

Model weights are published on Hugging Face; the model card lists the MIT license. The supplied model repository commit is dated October 25, 2025. This is the date of that commit, not necessarily the date of the first research announcement. (huggingface.co)

Pricing

As of November 17, 2026, official Glyph-specific API input and output prices could not be verified in either CNY or USD. The international pricing page did not list Glyph; the domestic pricing page could not be read without JavaScript. This does not establish that API use is free or that no price exists. (docs.z.ai)

Official sources