37 ranked

Took the throneQwen3.5 2Bvs Sep 15Aug · Qwen3.6-35B-A3B · heldSep · Qwen3.5 2BNations

Liberated

#	Model · Build	Hardware			Runtime	Proof
1NEW	Qwen3.5 2BQ4_K_M RTX 5090	RTX 5090	351	—	—	@fillagrew
2NEW	Swift 1.5 Qwen3.8 Flash-NextIQ2_XS RTX 4090	RTX 4090	220	3830	Strata	@Knuckles_XBT
3NEW	Qwen3.8-27BNVFP4 RTX A5000	RTX A5000	177.6	—	TensorFold	@Ja6ek
4NEW	Gemma-4 26B-A4B-itQ4_K_M RTX 5090	RTX 5090	173	—	—	@fillagrew
5NEW	Qwen3.8-Flash-NextIQ3_S RTX 5090	RTX 5090	142	—	Strata	@arianpg
6NEW	Qwen3.8-27BNVFP4 RTX A5000	RTX A5000	139.3	—	vLLM	@Ja6ek
7NEW	Qwen3.8-Flash-Next M5 Max	M5 Max	126.5	—	MTPLX V2.11.3	@Youssofal_
8NEW	Qwen3.8-Flash-NextIQ3_S RTX 5090	RTX 5090	104	1657	Strata	@DaSun64125381
9NEW	Qwen3.8-27B4-bit DGX Spark	DGX Spark	102.9	—	TensorFold	@Oluwaphilemon1
10▼9	Qwen3.6-35B-A3BNVFP4 DGX Spark	DGX Spark	94.4	181.5	—	@bonellisystems
11NEW	Qwen3.8-27B4-bit M5 Pro 64GB	M5 Pro 64GB	89.9	—	TensorFold	@aartiles24
12▼10	Qwen3.6-35B-A3B4bit M4 Pro	M4 Pro	85.5	—	rapid-mlx	@rapidmlx
13NEW	Qwen3.5-4B4bit M4 Pro	M4 Pro	82.8	—	rapid-mlx	@rapidmlx
14NEW	Qwen3.8-Flash-NextNVFP4 RTX 5090	RTX 5090	80.9	—	—	@tekizaihq
15NEW	Qwen3.8-Flash-NextEXL3 DGX Spark	DGX Spark	79.5	—	EXL3	@yume_arasaki
16NEW	Qwen3.8-Flash-NextNVFP4 DGX Spark	DGX Spark	74.8	—	TensorFold	@redp314
17NEW	Qwen3.8-27BNVFP4 DGX Spark	DGX Spark	72	—	—	@hasso5703
18NEW	Qwen3.8-Flash-NextIQ3_S RTX 3090 24GB	RTX 3090 24GB	72	—	Strata	@needmorevram
19NEW	AliceAI-Foundation-80B-A3BQ4_K_M M5 Max	M5 Max	65	—	llama.cpp	@aqty
20NEW	Qwen3.8-27B4-bit M6 mini 32GB	M6 mini 32GB	61.9	—	TensorFold	@jmurillocode
21NEW	GLM 5.3 FlashEXL3 DGX Spark	DGX Spark	60.4	1950	TensorFold	@MiaAI_lab
22NEW	GLM 5.3 FlashNVFP4 DGX Spark	DGX Spark	55.6	—	—	@PlusTen_AI
23NEW	Empero Qwen3.8-35B-A3BQ4_K_M RTX 3060	RTX 3060	50	550	llama.cpp	@Oluwaphilemon1
24NEW	Qwen3.5-9B4bit M4 Pro	M4 Pro	49.3	—	rapid-mlx	@rapidmlx
25▼22	Qwen3-8B4bit M4 Pro	M4 Pro	48.3	—	rapid-mlx	@rapidmlx
26NEW	MiMo-V2.6-Distill-Qwen-9BQ5_K_M RTX 5060	RTX 5060	47	1600	llama.cpp	@stfu0911
27NEW	Qwen3.8-Flash-NextFP8 DGX Spark	DGX Spark	45	—	—	@Oluwaphilemon1
28NEW	Qwen3.8-27BUD-Q4_K_XL RTX 4090	RTX 4090	40.7	2659.8	llama.cpp	@Oluwaphilemon1
29NEW	Ornith GTX 1660 SUPER	GTX 1660 SUPER	40	—	—	@mine_craft_bui
30▼26	Qwen3.6-35B-A3BUD-IQ3_XXS M1 Max 32GB	M1 Max 32GB	37	350	—	@moonsteroid
31NEW	Qwen3.8-Flash-NextIQ2_XS RTX 3060	RTX 3060	27.7	—	Strata	@dec21ai
32NEW	Qwen3.8-27BIQ3_XXS RTX 5060 Ti	RTX 5060 Ti	23	—	llama.cpp	@fntAInhead
33NEW	Qwen3.8-27BUD-Q2_K_XL RTX 5060 Ti	RTX 5060 Ti	19	—	llama.cpp	@fntAInhead
34NEW	GLM 5.3 FlashEXL3 DGX Spark	DGX Spark	18.6	—	EXL3	@vcruz305
35NEW	GLM 5.3 FlashNVFP4 DGX Spark	DGX Spark	15	—	vLLM	@sudoingX
36NEW	Nex-N2.5-miniQ4_K_M RTX 5060 Ti	RTX 5060 Ti	14	—	llama.cpp	@fntAInhead
37NEW	Qwen3.8-27B TurboFCFusionIQ2_M RTX 5060 Ti	RTX 5060 Ti	10	—	llama.cpp	@fntAInhead

Qwen3.8-Flash-Next

DGX Spark

DGX Spark · TensorFold · NVFP4 MTP-6

Pair

#16 / 37 in this filter

#4 / 10 on this chip

#6 / 9 of this model

Decode · tok/s

Prefill

Peak

On this chip

This model elsewhere

Lineage

Officialclaimed

Originating sentence

Qwen3.8-Flash-Next (NVFP4, MTP-6). Fixed 2,048-token code task: 74.8 tok/s single-stream decode

Flags

TensorFold 0.3.6.3 · MTP-6 · 2048-token code task

Caveat

Rank is the quoted 74.8 single-stream. Context sweep 56–65 skipped as a range. 50.8 at 260k stays in the note.

Proof

@redp314

Get

Get on TensorFold

Used boxes

1 listed DGX Sparkfrom $7,990 · street $8,050Market

Pairs

Liberated