Models / Qwen3.6 35B A3B

Qwen3.6 35B A3B

MoE enthusiast
HuggingFace → HuggingBay → Alibaba → Find for my hardware

35B total, 3B active per token (MoE: 256 experts, 8 routed + 1 shared) with hybrid Gated DeltaNet + Gated Attention layers. 256K native context, extensible to ~1M.

The agentic flagship - ranked #1 local model for agent workflows in 2026 (perfect tool-calling, zero catastrophic failures on an 84-scenario / 16-category / 8-trial benchmark).

  • Q4_K_M (~20GB) for GGUF runtimes.
  • Q8_0 (~37GB) for full precision.

AI-generated content marks

This model embeds text watermarks in generated text and adds C2PA provenance metadata to supported files such as .png, .jpg, and .svg. Marks can be lost through editing, screenshots, or format conversion, so their absence does not prove a file is human-made.

Provider transparency docs →

coding reasoning writing chat
Parameters
35.0B
Context
262k
License
apache 2.0
Developer
Alibaba
Origin
🇨🇳 China
Released
Apr 2026

Scores

Coding
90
Reasoning
91
Tool calling
95
General
87

Score per dollar

1740 pts per $/M input

general_score (87) divided by cheapest input price ($0.05/M). Higher is better value. See live pricing.

Related models

Qwen3.8-Omni-Flash
125.0B enthusiast no local build $0.15/M in Sep 14, 2026
Qwen3.8-Flash-Next
180.0B 6.0B active enthusiast 24GB min RAM Aug 26, 2026
Qwen3.8-27B
27.0B enthusiast 8GB min RAM Aug 5, 2026
Qwen3.8-Max
2400.0B 95.0B active premier no local build $2.00/M in Aug 3, 2026
Qwen3.6 27B
27.0B enthusiast 17GB min RAM $0.20/M in Nov 1, 2025
Qwen3 14B
14.7B consumer 9GB min RAM $0.10/M in Apr 28, 2025
Qwen3 235B A22B
235.0B 22.0B active workstation 68GB min RAM $0.46/M in Apr 28, 2025
Qwen3 30B A3B
30.5B 3.0B active enthusiast 16GB min RAM $0.12/M in Apr 28, 2025
Qwen3 32B
32.8B enthusiast 20GB min RAM $0.08/M in Apr 28, 2025
Qwen3 4B
4.0B edge 3GB min RAM Apr 28, 2025
Qwen3 8B
8.2B consumer 5GB min RAM $0.12/M in Apr 28, 2025

Guides covering Qwen3.6 35B A3B

DGX Spark benchmarks: which local model wins in 2026 Aug 19, 2026

Save your hardware and every model page answers the real question: will it run on your machine, and how fast?

Join free - save your rig →

Run it locally

Per-quant memory needs and a static "can you run it?" reference - no rig entry required

Q4_K_M
20.0GB 20.0GB min 22.0GB rec
Balanced - the usual local sweet spot
Q8_0
37.0GB 37.0GB min 40.0GB rec
Near-lossless

The reference hardware

Schematic of the NVIDIA Jetson Orin NX 16GB reference rig - 16GB unified memory, 102 GB/s aggregate bandwidth
Schematic of the Single GTX 1080 Ti (11GB) reference rig - 11GB VRAM, 484 GB/s aggregate bandwidth
Schematic of the 4x H100 80GB (320GB) reference rig - 320GB VRAM, 13400 GB/s aggregate bandwidth
Schematic of the NVIDIA DGX Station 748GB reference rig - 748GB unified memory, 8000 GB/s aggregate bandwidth
Schematic of the 8x RTX 3090 rack (192GB) reference rig - 192GB VRAM, 7489 GB/s aggregate bandwidth
Schematic of the 4x RTX 5090 (128GB) reference rig - 128GB VRAM, 7168 GB/s aggregate bandwidth
Schematic of the AMD Instinct MI300X (192GB) reference rig - 192GB VRAM, 5324 GB/s aggregate bandwidth
Schematic of the 4x RTX 4090 (96GB) reference rig - 96GB VRAM, 4032 GB/s aggregate bandwidth
Schematic of the 2x RTX 5090 (64GB) reference rig - 64GB VRAM, 3584 GB/s aggregate bandwidth
Schematic of the 2x RTX 3090 (48GB) reference rig - 48GB VRAM, 1872 GB/s aggregate bandwidth
Schematic of the Single RTX 5090 (32GB) reference rig - 32GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the RTX PRO 6000 Blackwell (96GB) reference rig - 96GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 192GB reference rig - 192GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 512GB reference rig - 512GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Single RTX 4090 (24GB) reference rig - 24GB VRAM, 1008 GB/s aggregate bandwidth
Schematic of the MacBook Pro M5 Max 128GB reference rig - 128GB unified memory, 614 GB/s aggregate bandwidth
Schematic of the Dual EPYC 9004 + 768GB DDR5-4800 reference rig - 768GB unified memory, 460 GB/s aggregate bandwidth
Schematic of the DGX Spark 128GB unified reference rig - 128GB unified memory, 273 GB/s aggregate bandwidth
Schematic of the Ryzen AI Max+ 395 128GB reference rig - 128GB unified memory, 256 GB/s aggregate bandwidth
Schematic of the Jetson AGX Orin 64GB reference rig - 64GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-3200 + 2x RTX 3090 reference rig - 560GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-2400 + 2x RTX 3090 reference rig - 560GB unified memory, 153 GB/s aggregate bandwidth

22 reference configs, drawn in-house. Scroll for more.

Can you run it? - reference rigs

Rig Q4_K_M Q8_0
NVIDIA Jetson Orin NX 16GB no -> cloud no -> cloud
Single GTX 1080 Ti (11GB) no -> cloud no -> cloud
4x H100 80GB (320GB) fast 4293.7t/s fast 2322.3t/s
NVIDIA DGX Station 748GB fast 2563.4t/s fast 1386.4t/s
8x RTX 3090 rack (192GB) fast 2399.8t/s fast 1298.0t/s
4x RTX 5090 (128GB) fast 2296.8t/s fast 1242.2t/s
AMD Instinct MI300X (192GB) fast 1706.2t/s fast 922.8t/s
4x RTX 4090 (96GB) fast 1291.9t/s fast 698.8t/s
2x RTX 5090 (64GB) fast 1148.4t/s fast 621.1t/s
2x RTX 3090 (48GB) fast 600.0t/s fast 324.5t/s
Single RTX 5090 (32GB) fast 574.2t/s offload
RTX PRO 6000 Blackwell (96GB) fast 574.2t/s fast 310.6t/s
Mac Studio M4 Ultra 192GB fast 381.7t/s fast 206.5t/s
Mac Studio M4 Ultra 512GB fast 381.7t/s fast 206.5t/s
Single RTX 4090 (24GB) fast 323.0t/s offload
MacBook Pro M5 Max 128GB fast 214.6t/s fast 116.1t/s
Dual EPYC 9004 + 768GB DDR5-4800 fast 147.7t/s fast 79.9t/s
DGX Spark 128GB unified fast 87.5t/s fast 47.3t/s
Ryzen AI Max+ 395 128GB fast 82.0t/s fast 44.4t/s
Jetson AGX Orin 64GB fast 65.6t/s fast 35.5t/s
Epyc + 512GB DDR4-3200 + 2x RTX 3090 fast 65.6t/s fast 35.5t/s
Epyc + 512GB DDR4-2400 + 2x RTX 3090 fast 49.2t/s fast 26.6t/s

Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.

Download options

Q4_K_M official -5% vs fp16
20.0GB dl 20.0GB min 22.0GB rec
REC RAM vs largest quant
Q8_0 official -1% vs fp16
37.0GB dl 37.0GB min 40.0GB rec
REC RAM vs largest quant

Or run it in the cloud

Live per-provider pricing, throughput and uptime - refreshed about 19 hours ago via OpenRouter. Click a column to sort.

Provider Type Input $/M Output $/M Cache $/M Tok/s Latency Uptime Value
Darkbloom
API 0.05 0.70 0.025 - - 100.00% best uptime
Venice
API 0.10 1.00 0.050 - - 100.00%
DekaLLM
API 0.10 1.00 0.050 - - 100.00%
AkashML
API 0.10 0.90 0.050 - - 100.00%
Parasail
API 0.10 0.95 0.050 - - 100.00%
Phala
API 0.20 1.27 0.056 - - 100.00%
SiliconFlow
API 0.24 1.80 0.150 - - 100.00%
CoreWeave
API 0.25 1.25 0.250 - - 100.00%
AtlasCloud
API 0.25 1.25 0.250 - - 100.00%
DeepInfra avoid
API 0.10 0.95 0.100 - - 50.00%

Default order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are "-". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.

Detailed API pricing page + JSON endpoint →

See who runs Alibaba in production →

PRICE HISTORY

Inference cost over time

Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.

Loading price history...