#1 GLM-5.3-Flash
Best capability that actually fits a 128-512 GB box, MIT license, mainstream llama.cpp support. The sweet spot for Spark owners.
- Version
- GLM-5.3-Flash
- Maker
- Z.ai (Zhipu)
- Release date
- 2026-08-25 (Hugging Face repo creation date unless noted)
- Architecture
- MoE, hybrid sparse+linear attention, native multimodal (image/video)
- Total params
- 320B
- Active params
- 18B
- Context length
- no published figure yet unverified
- License
- MIT
- Commercial notes
- Permissive; flagship GLM-5.3 (753B) uses a custom glm-5.3 license instead
- Variants
- GLM-5.3-Flash
- GLM-5.3-Flash-BF16
- GLM-5.3 (753B/~40B flagship)
- unsloth GGUF
- Memory
- BF16/native: ~306 GiB (Yotta Labs)
FP8: no published figure yet
Q8: no published figure yet
Q4: no published figure yet
LLMCheck: needs a 128 GB Mac at 3-bit in llama.cpp - Hardware
- Single Spark (128GB): Yes at ~3-bit (inferred from LLMCheck 128 GB Mac note) unverified
Dual Spark (256GB): Yes at ~4-bit (inferred) unverified
Quad Spark (~512GB): Yes, near-full precision fits (inferred) unverified
24GB GPU: No
128GB Mac / Strix: Yes at 3-bit (LLMCheck)
Spark rows inferred from published memory sizes, not tested - Tokens/sec
- no published figure yetSpeeds come from the cited sources. Entries marked "hosted, not local" are cloud API speeds, and "ESTIMATE" entries are estimates. Neither is a local measurement.
- Document Q&A strengths
- Agentic/tool use, outperforms GLM-5.3-Flash's predecessor GLM-5.2 per Z.ai
- Image+video input for scanned docs
- Top-4 open model on Artificial Analysis Intelligence Index (41.8)
- Embedding / reranker pairing
- no published figure yet
- Runtimes
- Ollama: listed (ollama.com/library/glm-5.3-flash; may be cloud tag) unverified
llama.cpp: Yes, merged upstream 2026-09-30 (b11279); MTP merged 2026-10-07
vLLM: Yes (official recipe)
LM Studio: no published figure yet unverified - Hugging Face
- https://huggingface.co/zai-org/GLM-5.3-Flash
DGX Spark configs it fits
- Single Spark (128GB): Yes at ~3-bit (inferred from LLMCheck 128 GB Mac note) unverified
- Dual Spark (256GB): Yes at ~4-bit (inferred) unverified
- Quad Spark (~512GB): Yes, near-full precision fits (inferred) unverified
Spark fit is estimated from published memory sizes, not tested on Spark hardware.
Sources
- https://huggingface.co/zai-org/GLM-5.3-Flash
- https://www.yottalabs.ai/post/best-chinese-llm-models-2026-deepseek-qwen-glm-kimi-compared
- https://llmcheck.net/blog/state-of-open-source-local-llms-october-2026/
- https://theopenweights.com/leaderboards/llm-intelligence
- https://ollama.com/library/glm-5.3-flash
Last updated Oct 24, 2026. Fields tagged unverified lack a primary source. "No published figure yet" means we found no source, and we don't guess. Spark fit is estimated from published memory sizes, not tested on Spark hardware. Ollama listings may be cloud-only tags; check this for Kimi K3 and DeepSeek V4.1 before pulling.
The RAGShelf weekly
New local RAG tools, releases, and tested setups. One email a week, no spam.