Advisor / Hardware / RTX PRO 6000 96GB (bonus page)

Best local LLMs for RTX PRO 6000 96GB (bonus page)

RTX PRO 6000 Blackwell Workstation. Memory-fit math with KV cache, speeds labelled by how they were obtained, and commands you can paste.

95.0 GBusable for model + KV cache
1792 GB/smemory bandwidth
56 GBRAM available for offload
6measured data points on file
Sponsor slot. No sponsor yet and no link.

Best for Coding · 16k context

ModelQuantFitMemorySpeed (generation)
GLM-4.5-AirQ5_K_Mfits fully86.6 GB~95 tok/s estimated
Gemma 4 31BQ8_0fits fully35.3 GB~33 tok/s estimated
Qwen3.6 27BQ8_0fits fully29.8 GB~39 tok/s estimated
gpt-oss 120BQ8_0fits fully64.0 GB~196 tok/s scaled from measured src
Qwen3-Coder 30B-A3BQ8_0fits fully34.1 GB~246 tok/s estimated
llama.cpp · top pick
llama-server -hf unsloth/GLM-4.5-Air-GGUF:Q5_K_M -c 16384 -fa on --jinja -ngl 99
ollama · top pick
OLLAMA_CONTEXT_LENGTH=16384 ollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q5_K_M

Best for Chat / general · 16k context

ModelQuantFitMemorySpeed (generation)
GLM-4.5-AirQ5_K_Mfits fully86.6 GB~95 tok/s estimated
Gemma 4 31BQ8_0fits fully35.3 GB~33 tok/s estimated
Qwen3.6 27BQ8_0fits fully29.8 GB~39 tok/s estimated
gpt-oss 120BQ8_0fits fully64.0 GB~196 tok/s scaled from measured src
Mistral Small 3.2 24BQ8_0fits fully27.7 GB~44 tok/s estimated
llama.cpp · top pick
llama-server -hf unsloth/GLM-4.5-Air-GGUF:Q5_K_M -c 16384 -fa on --jinja -ngl 99
ollama · top pick
OLLAMA_CONTEXT_LENGTH=16384 ollama run hf.co/unsloth/GLM-4.5-Air-GGUF:Q5_K_M

Best for Vision · 16k context

ModelQuantFitMemorySpeed (generation)
Gemma 4 31BQ8_0fits fully35.3 GB~33 tok/s estimated
Qwen3.6 27BQ8_0fits fully29.8 GB~39 tok/s estimated
Mistral Small 3.2 24BQ8_0fits fully27.7 GB~44 tok/s estimated
Gemma 4 12BQ8_0fits fully13.7 GB~44 tok/s scaled from measured src
Qwen3.6 35B-A3BQ8_0fits fully37.3 GB~281 tok/s estimated
llama.cpp · top pick
llama-server -hf unsloth/gemma-4-31B-it-GGUF:Q8_0 -c 16384 -fa on --jinja -ngl 99
ollama · top pick
OLLAMA_CONTEXT_LENGTH=16384 ollama run hf.co/unsloth/gemma-4-31B-it-GGUF:Q8_0

All 13 models that fit, fastest first · 16k context

Click a column header to sort. 6 of 13 speeds here are measured or scaled from measured; the rest are bandwidth estimates. Filter by use case, speed source, fit, license and more in the interactive advisor →

ModelParamsQuantFitMemoryTok/sSource
gpt-oss 20B20.91B MoEQ8_0fits fully12.5 GB~287scaled from measured
Qwen3.6 35B-A3B35.95B MoEQ8_0fits fully37.3 GB~281estimated
Qwen3-Coder 30B-A3B30.53B MoEQ8_0fits fully34.1 GB~246estimated
Gemma 4 26B-A4B25.81B MoEQ8_0fits fully27.9 GB~215estimated
gpt-oss 120B116.83B MoEQ8_0fits fully64.0 GB~196scaled from measured
GLM-4.5-Air110.47B MoEQ5_K_Mfits fully86.6 GB~95estimated
Qwen3.5 4B4.66BQ8_0fits fully5.1 GB~93scaled from measured
Qwen3-VL 8B8.77BQ8_0fits fully11.1 GB~66scaled from measured
Qwen3.5 9B9.65BQ8_0fits fully10.1 GB~60scaled from measured
Mistral Small 3.2 24B24.01BQ8_0fits fully27.7 GB~44estimated
Gemma 4 12B11.96BQ8_0fits fully13.7 GB~44scaled from measured
Qwen3.6 27B27.78BQ8_0fits fully29.8 GB~39estimated
Gemma 4 31B31.27BQ8_0fits fully35.3 GB~33estimated

Long context on this hardware · up to 1M tokens

Largest context whose KV cache still fits next to the weights (same placement as the 16k result, KV types f16 / q8_0 / q4_0). Models are capped at their verified maximum; extended means beyond the native maximum using YaRN, only listed where the model card documents it. Long context needs flash attention (-fa on, already in the commands) and a lot of memory; a quantized KV cache saves memory but can cost accuracy. Speed at depth is an estimate (penalty fitted to measured depth runs up to 128k).

Try it in the advisor: 128k · 256k · 512k · 1M

ModelNative maxMax ctx f16Max ctx q8_0Max ctx q4_0Est. tok/s at max f16 ctx
Qwen3.6 35B-A3B Q8_0256k1M extended1M extended1M extended~49 estimated
Qwen3.5 4B Q8_0256k1M extended1M extended1M extended~15 estimated
Qwen3.5 9B Q8_0256k1M extended1M extended1M extended~17 estimated
Qwen3.6 27B Q8_0256k1M extended1M extended1M extended~15 estimated
Qwen3-Coder 30B-A3B Q8_0256k512k extended1M extended1M extended~21 estimated
Qwen3-VL 8B Q8_0256k512k extended1M extended1M extended~9 estimated
Gemma 4 26B-A4B Q8_0256k256k256k256k~108 estimated
Gemma 4 12B Q8_0256k256k256k256k~35 estimated
Gemma 4 31B Q8_0256k256k256k256k~22 estimated
gpt-oss 20B Q8_0128k128k128k128k~136 estimated
gpt-oss 120B Q8_0128k128k128k128k~87 estimated
Mistral Small 3.2 24B Q8_0128k128k128k128k~27 estimated
GLM-4.5-Air Q5_K_M128k32k64k128k~64 estimated
Affiliate box: “cheapest upgrade” for RTX PRO 6000 96GB (bonus page). No affiliate links yet.

Open the interactive advisor with this hardware →

Speeds marked measured are medians of real runs (short-context generation) from these sources; scaled from measured means a measurement of a same-size or same-model-different-quant run, scaled by bytes read; estimated comes from memory bandwidth (spec source: https://www.techpowerup.com/gpu-specs/rtx-pro-6000-blackwell.c4272) × an efficiency factor of 0.61 calibrated to measured runs on this hardware. 6 measured data points on file for this hardware · submit your benchmark. System RAM here is an example (64 GB, ddr5-5600). Model data last verified 2026-10-14. Ranking = fits and fast enough first, then larger; not a quality benchmark.

Page generated 2026-10-16 by tools/generate_pages.js