GLM-4.7-Flash

GLM-4.7-Flash was released on 19 January 2026 as the free-tier companion to GLM-4.7, the model that closed out Z.ai's GLM-4 generation a month earlier. It is a mixture-of-experts model with 31.2 billion parameters in total: 47 layers, 64 routed experts plus one shared, four of them active per token, and a context window of 202,752 tokens. The weights are on Hugging Face under the MIT licence, and calls through the Z.ai API cost nothing — input, output and cached input are all listed as free. The interesting number is not the benchmark score, it is the download counter. In a thirty-day window GLM-4.7-Flash was pulled about 1.88 million times, against roughly 65,000 for the full GLM-4.7 it descends from — a factor of about twenty-nine. Only the company's OCR model is downloaded more. This is the clearest evidence in Z.ai's catalogue of a pattern the whole open-weights market keeps demonstrating: what decides adoption is whether the file fits on the hardware someone already owns, not where the model sits in a leaderboard. At 31 billion parameters with four experts active per token, it is small enough to serve from a single well-equipped machine, and Z.ai positions it for the work where volume matters more than peak quality: high-frequency coding assistance, drafting, translation, long-form writing and role play, all at low latency. The vendor describes it as competitive in coding at its own scale — a careful phrase, and the right one, since nothing here challenges the 358-billion-parameter flagship on hard reasoning. The practical consequence is worth stating plainly for anyone choosing between the two. GLM-4.7 costs $0.60 per million input tokens and $2.20 per million output. GLM-4.7-Flash costs nothing at all through the same API, keeps essentially the same context window, and can additionally be downloaded and run privately. The flagship earns its price on the hardest tasks; for everything else, the free model is the one the market has actually picked.

▸Videos

No videos yet.

▸Reviews