Can you run Qwen3.5 35B-A3B Instruct on RTX 5090?

Qwen3.5 35B-A3B Instruct Q4_K_M on the NVIDIA GeForce RTX 5090: verdict, VRAM math and estimated speed.

Yes, it runs
Quick answer

Yes, the NVIDIA GeForce RTX 5090 runs Qwen3.5 35B-A3B Instruct. 20 GB weights at Q4_K_M vs 28.8 GB usable VRAM; ~118 tok/s est. (ModelFit, 2026).

$ollama run qwen3.5:35b-a3b
VERDICT
Comfortable
EST. SPEED
~118 tok/s
WEIGHTS
20 GB Q4_K_M

VRAM math and speed are ModelFit engine estimates, not measurements. Commands are registry-verified Ollama tags.

Cite this page: ModelFit, Qwen3.5 35B-A3B Instruct on RTX 5090, https://modelfit.io/can-i-run/qwen3.5-35b-a3b-q4-on-rtx-5090/, updated September 2026, CC BY 4.0.

Last updated: September 18, 2026 · Editor: ModelFit Team

VRAM
32 GB (28.8 usable)
Model weights
20 GB Q4_K_M
Est. speed
~118 tok/s
First token
~0.9s
Fit grade
A · 31

What limits this combo

The four constraints the fit engine checks for Qwen3.5 35B-A3B Instruct on the RTX 5090, in order of what usually breaks first.

Weights fitOK20 GB of weights against 28.8 GB usable — 8.8 GB of headroom for context and the runtime.
Context ceilingOKWeights + KV cache stay in budget up to 128k tokens (22.5 GB total). Quantizing the KV cache to q8_0 roughly halves the KV column and pushes this further.
SpeedOK~118 tok/s (est.) on 1792 GB/s of memory bandwidth — comfortable for interactive chat and agentic loops.
Use-case ceilingOKStrongest fit: Chat (A). Agentic coding clears its ~30 tok/s floor, so tool-calling loops stay usable.

Quant explorer

Switch between the quality-gated builds tracked for this exact combo. Q4_K_M stays the default recommendation; heavier Q6/Q8 builds only appear when their exact Ollama tags are registry-verified.

ModelFit engine estimate
Verdict
Runs
Est. speed
~118 tok/s
First token
~0.9s · Instant
Safe context
128k
A · 31 Excellent headroom

Why no 2-bit builds: Q1/Q2-class quants can look attractive in a memory table, but their quality loss is large enough that ModelFit excludes them from rankings instead of inflating the catalog with junk options. Usable VRAM: 28.8 GB.

Workload verdicts

Qwen3.5 35B-A3B Instruct on the RTX 5090, graded per use case from the model's registry-verified tuning and the engine's speed estimate for this exact combo.

ModelFit engine estimate
ChatA~118 tok/s est. vs ~20 needed for chat.
CodingA~118 tok/s est. vs ~15 needed for coding.
Agentic codingA~118 tok/s est. vs ~30 needed for agentic coding.
ReasoningA~118 tok/s est. vs ~12 needed for reasoning.
RAGA~118 tok/s est. vs ~20 needed for rag.

Grades combine tag-verified tuning (a model not built for the workload caps at C) with the tok/s floor each workload needs to feel usable. A combo that partially offloads caps at C; one that does not fit is D everywhere.

Memory math: weights + context vs budget

Weights take 20 GB. Context costs extra KV-cache on top — this is where long-context sessions break on cards that technically fit the weights.

Weights 20 GBKV cache · 16k context 0.3 GBHeadroom 8.5 GBRuntime reserve 3.2 GB

The model leaves about 8.5 GB of the usable VRAM budget free after weights and 16k context.

ContextKV-cacheTotalFits
8k tokens0.2 GB20.2 GBFits
16k tokens0.3 GB20.3 GBFits
32k tokens0.6 GB20.6 GBFits
64k tokens1.3 GB21.3 GBFits
128k tokens2.5 GB22.5 GBFits

KV-cache figures assume an fp16 cache, the llama.cpp/Ollama default. Standard GQA models use a size-class estimate (8 KV heads x 128 head dim class); hybrid linear-attention models (Qwen3.5/3.6, Qwen3-Next) use the exact per-token cost from their published config, since only their sparse full-attention layers cache KV. A q8_0 KV cache roughly halves either figure. Estimates, not measurements.

See how fast it feels

A deterministic typing simulation for Qwen3.5 35B-A3B Instruct: first token ~0.9s (instant prefill), then ~118 tokens/sec.

Start the simulation to preview the response pace.

ModelFit engine estimate, not a measured benchmark. Real speed varies with prompt length, thermals, runtime, and KV-cache settings.

Qwen3.5 35B-A3B Instruct on RTX 5090: FAQ

Can the NVIDIA GeForce RTX 5090 run Qwen3.5 35B-A3B Instruct?

Yes. Qwen3.5 35B-A3B Instruct (Q4_K_M) loads in about 20 GB and the RTX 5090 offers 28.8 GB of usable VRAM, leaving headroom for context. Expect at roughly 118 tokens/sec (est.).

How much VRAM does Qwen3.5 35B-A3B Instruct need?

About 20 GB for the weights at Q4_K_M, plus KV-cache for context: roughly 0.3 GB extra at 16k tokens. The RTX 5090 budget is 28.8 GB (32 GB x 90%).

What is the best quantization of Qwen3.5 35B-A3B Instruct for the RTX 5090?

Stick with the Q4_K_M build at 20 GB — every heavier quant exceeds the 28.8 GB usable VRAM.

What GPU do I need to run Qwen3.5 35B-A3B Instruct comfortably?

The RTX 5090 already runs Qwen3.5 35B-A3B Instruct comfortably. Larger cards only buy you longer context or a heavier quant.

How much context can Qwen3.5 35B-A3B Instruct use on the RTX 5090?

Up to 128k tokens stay fully in VRAM (22.5 GB total with the KV cache). Quantizing the KV cache to q8_0 roughly halves the KV column and buys back about one context tier.

Is Qwen3.5 35B-A3B Instruct on the RTX 5090 fast enough for coding agents?

Yes: the engine estimates ~118 tok/s against the ~30 tok/s an agentic loop needs to feel responsive, and Qwen3.5 35B-A3B Instruct is tuned for tool-calling.