GLM-4.7 is the GLM model that made programming the family's headline capability. It shipped enhanced code generation, stable multi-step reasoning, better handling of complex agent tasks, more natural dialogue and visibly improved front-end aesthetics — the last of which matters more than it sounds, because a model that writes plausible interface code saves a category of manual work that pure back-end benchmarks never capture.
It sits directly before the 5 series in the lineage, and much of what the 5 series is praised for started as a 4.7 improvement. It is also the generation that produced the 30B GLM-4.7-Flash tier, which remains the family's answer for teams that need a genuinely small model without leaving the same MIT license and the same weight hosts. Read together, the two 4.7 releases are the point where the family stopped being an interesting research line and became something you could reasonably put behind a product.
No sign-up, no account, no credits. Type a prompt and get a streamed answer.
Response will stream here...
This playground runs a free open-source LLM via OpenRouter. It is not GLM-5.2 and does not use GLM weights — it's here so you can test a live model while you read the published GLM benchmark data below.
Most coding evaluations score whether a patch makes a test pass. That misses an entire class of work: producing interface code a human will accept without rewriting it. Layout that holds together, sensible spacing, components that follow the conventions of the framework in use — these are judged visually, not by a test runner, and a model that gets them wrong generates work rather than saving it. GLM-4.7's improvement there, alongside its more natural dialogue, made it usable as a day-to-day pair for application work rather than only for isolated algorithmic tasks. The same generation also tightened stable multi-step reasoning, which is what allowed the GLM-5 line to concentrate on horizon length instead of basic execution reliability.
The published comparison table covers GLM-5.2 and GLM-5.1. The column below is GLM-5.1's — the closest published reference point for the generation that followed GLM-4.7. No benchmark column has been published for GLM-4.7 itself.
No GLM-4.7 row exists in the published comparison set, so none is quoted here. The reference column shows what the next generation achieved on coding: 58.4 on SWE-bench Pro, 50.9 on ProgramBench and 42.7 on NL2Repo for GLM-5.1, against 62.1, 63.7 and 48.9 for GLM-5.2. If you need measured programming performance rather than positioning, those are the two columns worth reading.
Comparison on ProgramBench, as published by the model authors.
ProgramBench is the published evaluation closest to GLM-4.7's stated strength. The frontier leads — Claude Opus 4.8 at 71.9 and GPT-5.5 at 70.8 — with GLM-5.2 at 63.7, GLM-5.1 at 50.9 and DeepSeek-V4-Pro at 47.8. The open-weight field spans more than fifteen points on this row, which is a useful reminder that 'open model' is not a performance tier: the specific release you pick matters far more than the licence it carries.
GLM-4.7 follows GLM-4.5-Air and precedes GLM-5. Its own lightweight sibling, GLM-4.7-Flash, is a 30B model from the same generation, and the whole line converges on GLM-5.2 today. If you are picking a GLM model for programming work now, the flagship carries every 4.7-generation improvement plus the 1M context window and a published benchmark column. The useful way to think about the lineage is that capability accumulated while the licence stayed constant: everything from the 4.5 tier upward is MIT with weights on the same public hosts, so moving up the family is a compute decision and never a legal one.
The 1M-token context flagship. It is the open-source state of the art on coding and long-horizon evaluations, with explicit Non-Thinking, High and Max effort levels.
The lightweight 30B tier. Efficient and fast enough for local inference, and reported to outrank open models of a similar scale.
Read the GLM model benchmarks, then put a real model to work. The playground above is free and needs nothing from you; if you want a fuller AI toolkit, our partner's free tier starts here.