Up to 96 GB RAM is a bummer. 128 GB RAM would fit IQ3_XXS quant of huggingface.co/unsloth/DeepSeek-V4-Flash-GGUF (103 GB). (a much improved, final version of DeepSeek-V4-Flash is expected this month)
See DeepSeek-V4-Flash and various other, up to 300B, MoE and dense LLM models evaluated:
artificialanalysis.ai/?models=qwen3-5-122b-a10b-non-reasoning%2Cqwen3-5-122b-a10b%2Cmimo-v2-5-0424%2Cqwen3-6-27b-non-reasoning%2Cminimax-m2-7%2Cqwen3-6-35b-a3b-non-reasoning%2Cstep-3-7-flash%2Cqwen3-6-35b-a3b%2Cdeepseek-v4-flash-high%2Cqwen3-6-27b%2Cdeepseek-v4-flash-non-reasoning%2Chy3%2Cgemma-4-31b%2Cgemma-4-31b-non-reasoning%2Cdeepseek-v4-flash%2Cmistral-medium-3-5%2Cgpt-oss-120b%2Cnvidia-nemotron-3-super-120b-a12b&intelligence=artificial-analysis-intelligence-index
But what laptop supports up to 128 GB RAM or even the full 256 GB RAM, you may ask? FRAMEWORK (but may be also CLEVO and TONGFANG):
community.frame.work/t/status-report-fw13-amd-7840-debian-trixie-hibernation-128-gb-ram-etc/69864
reddit.com/r/framework/comments/1rglnz1/128gb_memory_on_the_framework_16_laptop/
(type "128 gb" or "256 gb" into each' search)
The IQ3_XXS quant is good, see: huggingface.co/unsloth/DeepSeek-V4-Flash-GGUF/discussions/6
Recently also several commits to llama.cpp regarding DeepSeek-V4-Flash have landed (so now prompt processing speeds are almost doubled, KV-cache size is much smaller, and decode speed is much faster, too):
33a75f41c 2026-07-15 DeepseekV4: reduce graph splits (#25702)
7f575c39d 2026-07-14 DeepseekV4: fix seq_rm (#25588)
13f2b28b0 2026-07-11 DeepseekV4: clear cache only for seq rather than full (#25521)
00f5442cc 2026-07-11 ggml : add GGML_OP_LIGHTNING_INDEXER that implements DeepSeek V3.2/V4 lightning indexer (#24231)
2ed3c1abb 2026-07-10 llama : make all KQ masks f16 if FA is used, remove zero attention bias, remove raw_k repeats in DeepSeek V4 (#25370)
8c146a836 2026-06-29 DeepSeek V4 (#24162)