About GLM-5.2
GLM-5.2 is Z.ai's flagship open release, published under MIT with the card's pointed phrase — no regional limits, technical access without borders. It is a very large mixture-of-experts system with a small active fraction per token, built for long-horizon agentic engineering: sustained coding sessions, tool use and multi-step tasks across a stable million-token context, with selectable thinking-effort levels.
Its sparse-attention design shares indexing work across layers specifically to keep very long contexts computationally affordable, which is the engineering behind the long-horizon positioning.
GLM-5.2 is Z.ai's 753B-parameter model released in June 2026, with a 1024K-token context window. It is a mixture-of-experts model: all 753B parameters must sit in memory, but only ~38B are active per token, which is what makes it fast for its size. It uses a latent-attention (MLA) design that compresses its KV cache, keeping even very long contexts cheap in memory: its KV cache is about 0.7 GB at an 8K context, 11.8 GB at 128K, and 94.2 GB at the full 1024K window (FP16 cache).
Scale first: the resident expert pool places every quantisation in the totals above beyond single-device territory — this is a multi-accelerator server model, or the outer limit of the largest unified-memory configurations at the most aggressive quantisation. Once housed, the design is considerate: compressed attention keeps the million-token cache remarkably small for the context it covers, and the modest active set keeps per-token compute reasonable for the size. Z.ai maintains first-party support across the major serving stacks, and the MIT licence removes every legal question self-hosters usually have to check.
For most people Q4_K_M is the sweet spot — the most popular quality/size trade-off — while Q8 is near-lossless if you have the memory. Totals above include the KV cache and a realistic framework overhead, so they are what you should expect to see in practice rather than just the download size. Weight sizes are calibrated against real GGUF files — see the methodology.
Frequently asked questions
How much VRAM does GLM-5.2 need?
At Q4_K_M with an 8K context, GLM-5.2 needs about 507 GB (weights 459 GB + KV cache + overhead). No single consumer or workstation device fits it — it needs multi-GPU or datacenter hardware.
Can an RTX 4090 (24GB) run GLM-5.2?
Not fully in VRAM. GLM-5.2 needs about 507 GB even at Q4_K_M, so a 24 GB card would have to offload layers to system RAM at a large speed penalty.
Can a Mac run GLM-5.2?
Not on any current configuration — even a 512 GB Mac Studio's ~384 GB GPU allowance is below the ~507 GB this model needs.
How much VRAM does GLM-5.2 need at its full 1024K context?
The KV cache grows with context: about 0.7 GB at 8K tokens versus 94.2 GB at the full 1024K window (FP16 cache). Add that to the weights (459 GB at Q4_K_M) plus overhead — long contexts can cost more than a whole quantisation step.
Related
VRAM calculator for any model ·Token counter
Similar-size local models: Mistral Large 3 (675B) · Llama 4 Maverick (400B) · Qwen3.5 397B-A17B (397B) · DeepSeek V4 Flash (284B) · Mistral Small 4 (119B)
Last updated 2026-09-03. Architecture figures from the model's published config.json; see the methodology.