Qwen3.6 35B A3B · 7 GPUs measured first-party · llama.cpp · Updated October 2026
What GPU Do You Need for Qwen3.6 35B A3B?
Qwen3.6 35B A3B on 7 GPUs, measured first-party: NVIDIA GeForce RTX 5090 leads at 300.1 tok/s, L4 trails at 79 tok/s, and it peaked at 22GB of VRAM.
Benchmarked weights: bartowski/Qwen_Qwen3.6-35B-A3B-GGUF
NVIDIA GeForce RTX 5090
300.1 tok/s on Qwen3.6 35B A3B, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.
- 300.1 tok/s on Qwen3.6 35B A3B
- 32GB, clears the Qwen3.6 35B A3B floor
- 575W board rating
NVIDIA RTX PRO 6000 Blackwell Workstation Edition
258.7 tok/s on Qwen3.6 35B A3B, lowest launch price that still fits. Measured on our bench. 96GB of VRAM, $8,565 at launch.
- 258.7 tok/s on Qwen3.6 35B A3B
- 96GB, clears the Qwen3.6 35B A3B floor
- Rentable by the hour rather than bought
- 600W board rating
- Datacenter or workstation hardware, not a retail purchase
What GPU Do You Need for Qwen3.6 35B A3B?, tok/s by GPU
Efficiency: tok/s per 100W drawn
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: tok/s per $1,000 of MSRP
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
Qwen3.6 35B A3B. Measured tokens per second by GPU
| GPU | tok/s | Prompt t/s | tok/W | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 300.1 | 9462.8 | 1.43 | 209.8 W |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 258.7 | 8159.2 | 1.76 | 146.9 W |
| NVIDIA H100 80GB HBM3 | 236.7 | 5886.6 | 1.46 | 162.6 W |
| NVIDIA L40S | 171.2 | 6014.6 | 1.23 | 139.5 W |
| NVIDIA A100 80GB SXM4 | 149 | 3166.4 | 1.17 | 127.2 W |
| NVIDIA A10G | 117.9 | 2198.6 | 1.17 | 100.6 W |
| NVIDIA L4 | 78.98 | 1886.9 | 1.56 | 50.5 W |
What the numbers show. Across 7 GPUs measured on our own bench, RTX 5090 is fastest at 300 tok/s. The slowest, L4, manages 79.0, so the spread is 3.8x from top to bottom. RTX PRO 6000 Blackwell Workstation Edition is the most efficient, 259 tok/s at 147W.
About Qwen3.6 35B A3B. Qwen3.6 35B A3B: from Qwen, 36B parameters, on Hugging Face since April 2026, Apache 2.0 licence. 9,347,547 downloads in the last 30 days and 2 community quantizations.
How it compares. H100 80GB HBM3: Qwen3.6 35B A3B 236.7 tok/s, Ornith 1.5 35B A3B 240.2 (36B), Dolphin 2.9.1 Yi 1.5 34B 76.76 (34B), Qwen3-32B 74.07 (33B), DeepSeek-R1 Distill 32B 75.8 (33B). 1 of 4 beat Qwen3.6 35B A3B here.
Cost on a rented GPU. 1M generated tokens of Qwen3.6 35B A3B: $0.36 on a RTX 5090 ($0.39/hr, 56 min).
Qwen3.6 35B A3B: cost per 1M generated tokens on rented GPUs
| GPU | Cheapest rate | Speed (tok/s) | Cost per 1M generated tokens |
|---|---|---|---|
| NVIDIA GeForce RTX 5090 | $0.39/hr | 300.1 | $0.36 |
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | $1.08/hr | 258.7 | $1.16 |
| NVIDIA L40S | $0.79/hr | 171.2 | $1.28 |
| NVIDIA L4 | $0.44/hr | 78.98 | $1.55 |
| NVIDIA A100 80GB SXM4 | $0.95/hr | 149 | $1.77 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 236.7 | $2.51 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for Qwen3.6 35B A3B. 30+ tok/s: 7 (RTX 5090). 30 tok/s is roughly where replies outpace reading.
Reading your prompt. Before Qwen3.6 35B A3B writes anything it reads the input: 9462.8 tok/s on the RTX 5090 (0.4s for a 4,000-token prompt), 1886.9 on the L4 (2.1s). Long documents and big code files feel this number more than the generation speed.
VRAM for Qwen3.6 35B A3B. Measured peak 20.4GB, so 24GB is the smallest common card size; smallest card it ran on: A10G (24GB). With long context: Q4_K_M 26GB (tested), Q2_K 16GB, Q3_K_M 20GB, Q5_K_M 31GB, Q6_K 37GB.
Power on Qwen3.6 35B A3B. Most efficient: RTX PRO 6000 Blackwell Workstation Edition, 147W, 0.16 kWh per 1M generated tokens. Hungriest: RTX 5090, 210W, 0.19 kWh. At $0.15/kWh: $0.024 per 1M generated tokens.
Our verdict
Fastest on Qwen3.6 35B A3B: NVIDIA GeForce RTX 5090, 300.1 tok/s. Cheapest to rent per job: NVIDIA GeForce RTX 5090, $0.36 per 1M generated tokens.
FAQ
- What GPU do I need to run Qwen3.6 35B A3B?
- About 21GB. Cheapest consumer card that ran it: NVIDIA GeForce RTX 5090 (32GB, 300.1 tok/s).
- How fast is Qwen3.6 35B A3B on the NVIDIA GeForce RTX 5090?
- 300.1 tok/s at 210W, faster than every datacenter card we ran it on.
- How much does it cost to run Qwen3.6 35B A3B in the cloud?
- $0.36 per 1M generated tokens on a NVIDIA GeForce RTX 5090 at $0.39/hr, cheapest of 6 rentable cards we measured.
- Can I run Qwen3.6 35B A3B on a 12GB, 16GB or 24GB card?
- It used 20.4GB at the precision we tested. 12GB: no; 16GB: only at Q2_K (16GB); 24GB: yes.
- Is the RTX 5090 or the H100 80GB HBM3 faster for Qwen3.6 35B A3B?
- The RTX 5090: 300.1 vs 236.7 tok/s, 27% faster on our bench.
- Should I buy or rent a RTX 5090 for Qwen3.6 35B A3B?
- Its $1,999 launch price buys 5,139 rented hours at $0.39/hr, enough for about 5,552x 1M generated tokens of Qwen3.6 35B A3B. Buy only if you'll run more than that.
How we test
llama.cpp llama-bench at Q4_K_M, 512-token prompt and 128 generated tokens, three runs after a warmup, full GPU offload, with power and VRAM sampled throughout. Token generation is memory-bandwidth-bound, so the ranking tracks bandwidth closely, which makes it a fair guide to cards we haven't run yet.