Latest26 Aug 2026Open-weights LLM
GLM-5.3-Flash
A smaller, cheaper GLM-5.3 and the first natively multimodal model in the GLM-5 series. It can look at interfaces and rendered output, closing the loop between code, browsers and GUIs. Z.ai says it approaches top closed models on coding and agent work at about a tenth of GLM-5.2's price.
- 320B total / 18B active
- 300K context with context management
- Image and text input, text output
- Hybrid sparse and linear attention
AccessZ.ai API; open weights on Hugging Face, MIT licence
Price$0.15 input / $0.50 output per 1M tokens; cached input $0.03
14 Aug 2026Frontier LLM
GLM-5.3
Z.ai's current flagship, a post-training upgrade of GLM-5.2 aimed at coding and agents. Z.ai reports a 50% gain over GLM-5.2 on its own coding benchmark and new strength at finding software vulnerabilities, which led it to delay the weights for a safety review.
- 1M-token context; 128K max output
- Text-only input
- Same base model as GLM-5.2 (lab docs)
- CyberGym 84.5, up from 77.2 for GLM-5.2 (lab-reported)
- Reasoning effort: low, high, max
AccessGLM Coding Plan and Z.ai API first; open weights later on Hugging Face and ModelScope under the GLM-5.3 License
Price$1.40 input / $4.40 output per 1M tokens; cached input $0.26
16 Jun 2026Open-weights LLM
GLM-5.2
The GLM flagship from June 2026, centred on a usable 1M-token context so agents lose track of their goals less often in long tasks. It added flexible thinking-effort levels for trading quality against speed, and its weights were released under the MIT licence with no regional restrictions.
- 1M-token context
- Sparse attention with IndexShare (2.9x fewer FLOPs at 1M, lab claim)
- SWE-bench Pro 62.1 (lab model card)
- Multiple thinking-effort levels
AccessZ.ai API and chat; open weights on Hugging Face, MIT licence
Price$1.40 input / $4.40 output per 1M tokens; cached input $0.26
7 Apr 2026Open-weights LLM
GLM-5.1
An agentic-engineering upgrade of GLM-5 that Z.ai says can work on its own for up to eight hours in a single run, revising its plan over hundreds of rounds and thousands of tool calls. It brought significantly stronger coding and was released as MIT-licensed open weights.
- Up to 8 hours autonomous work (lab claim)
- MoE with DSA sparse attention (lab model card)
- SWE-Bench Pro 58.4 (lab model card)
- 128K max output (Alibaba Model Studio listing)
AccessZ.ai API, chat.z.ai; open weights on Hugging Face, MIT licence
Price$1.40 input / $4.40 output per 1M tokens; cached input $0.26
12 Feb 2026Open-weights LLM
GLM-5
Z.ai's fifth-generation flagship, built for complex systems engineering and long agent tasks rather than single coding prompts. It was pre-trained on 28.5T tokens, up from 23T for GLM-4.5, and adopted DeepSeek Sparse Attention to cut serving costs. Z.ai benchmarked it against Claude Opus 4.5.
- 744B total / 40B active
- 28.5T pre-training tokens
- About 200K context (202,752 tokens)
- SWE-bench Verified 77.8 (lab model card)
AccessZ.ai API, chat.z.ai; open weights on Hugging Face, MIT licence
Price$1.00 input / $3.20 output per 1M tokens; cached input $0.20
14 Jan 2026Image model
GLM-Image
An open image generator and editor that pairs an autoregressive model for understanding the prompt with a diffusion decoder for detail. It is aimed at images dense with text and knowledge, such as posters and diagrams, and handles both text-to-image and editing in one model.
- 9B autoregressive generator plus 7B diffusion decoder
- Text-to-image and image editing in one model
- Glyph encoder for text rendering
- Identity-preserving and multi-subject editing
AccessZ.ai API; open weights on Hugging Face, MIT licence (some components Apache-2.0)
Price$0.015 per image
22 Dec 2025Open-weights LLM
GLM-4.7
The last GLM-4 generation flagship, tuned as a coding partner. It added thinking that is kept across turns of a conversation and per-turn control over reasoning, with gains in multilingual coding, terminal tasks and web interface generation over GLM-4.6. Weights are MIT-licensed.
- Hugging Face lists 358B parameters
- SWE-bench Verified 73.8 (lab model card)
- Interleaved and preserved thinking
- Terminal Bench 2.0: 41 (lab model card)
AccessZ.ai API, GLM Coding Plan; open weights on Hugging Face, MIT licence
Price$0.60 input / $2.20 output per 1M tokens; cached input $0.11
8 Dec 2025Open-weights LLM
GLM-4.6V (and GLM-4.6V-Flash)
Z.ai's open vision-language models in two sizes, a 106B cloud model and a 9B local one. They can pass images straight into tool calls, read long documents with charts and tables, and turn UI screenshots into HTML and CSS that can then be edited in plain language.
- GLM-4.6V: 106B; GLM-4.6V-Flash: 9B
- 128K context
- Native vision-driven tool use
- Screenshot-to-front-end code
AccessZ.ai API; open weights on Hugging Face, MIT licence
PriceGLM-4.6V: $0.30 input / $0.90 output per 1M tokens; cached input $0.05