Pairs

69 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	351	—	—	@fillagrew
2NEW	Qwen3.8-27BNVFP4	RTX PRO 6000 Blackwell Max-Q	272.8	—	TensorFold	@llm_frontier_jp
3NEW	GLM 5.3 FlashEXL3	4× RTX PRO 6000	264	—	TensorFold 0.6.2	@Kurcide
4NEW	GLM 5.3 FlashEXL3	2× RTX PRO 6000	238	—	TensorFold 0.6.5	@Kurcide
5NEW	Swift 1.5 Qwen3.8 Flash-NextIQ2_XS	RTX 4090	220	3830	Strata	@Knuckles_XBT
6NEW	Qwen3.8-27BNVFP4	RTX A5000	177.6	—	TensorFold	@Ja6ek
7NEW	Gemma-4 26B-A4B-itUncensoredQ4_K_M	RTX 5090	173	—	—	@fillagrew
8NEW	Qwen3.8-27BNVFP4	RTX PRO 6000 Blackwell Max-Q	151.4	—	vLLM	@llm_frontier_jp
9NEW	Qwen3.8-Flash-NextIQ3_S	RTX 5090	142	—	Strata	@arianpg
10NEW	Qwen3.8-27BNVFP4	RTX A5000	139.3	—	vLLM	@Ja6ek
11NEW	Qwen3.8-Flash-Next	M5 Max	126.5	—	MTPLX V2.11.3	@Youssofal_
12NEW	GLM 5.3 FlashEXL3 4bpw	RTX PRO 6000 + 2× GB10	120	—	TensorFold	@hokazuya
13▼12	Qwen3.6-35B-A3B4-bit	M4 Pro 64GB	108.6	—	Rapid-MLX	@draslan_eth
14NEW	Qwen3.8-Flash-NextIQ3_S	RTX 5090	104	1657	Strata	@DaSun64125381
15NEW	Qwen3.8-27B4-bit	DGX Spark	102.9	—	TensorFold	@Oluwaphilemon1
16NEW	Qwen3.8-Flash-NextIQ2_XS	RTX 3090 24GB	100.6	—	Strata	@draslan_eth
17NEW	Qwen3.8-Flash-NextIQ3_S	2× RTX 5060 Ti 16GB	99.9	—	Strata	@needmorevram
18▼16	Qwen3.6-35B-A3BUncensoredNVFP4	DGX Spark	94.4	181.5	—	@bonellisystems
19NEW	Qwen3.8-Flash-NextQ4_XL	RTX PRO 6000	92.8	—	llama.cpp b11412	@piyo621
20NEW	Qwen3.8-27B4-bit	M5 Pro 64GB	89.9	—	TensorFold	@aartiles24
21▼18	Qwen3.6-35B-A3B4bit	M4 Pro	85.5	—	rapid-mlx	@rapidmlx
22NEW	GLM 5.3 FlashEXL3	DGX Spark	85	—	TensorFold	@MehdiHacks
23NEW	Qwen3.5-4B4bit	M4 Pro	82.8	—	rapid-mlx	@rapidmlx
24NEW	Qwen3.8-Flash-NextNVFP4	RTX 5090	80.9	—	—	@tekizaihq
25NEW	Qwen3.8-Flash-NextEXL3	DGX Spark	79.5	—	EXL3	@yume_arasaki
26NEW	Qwen3.8-Flash-NextNVFP4	DGX Spark	74.8	—	TensorFold	@redp314
27NEW	Qwen3.8-27BNVFP4	DGX Spark	72	—	—	@hasso5703
28NEW	Qwen3.8-Flash-NextIQ3_S	RTX 3090 24GB	72	—	Strata	@needmorevram
29NEW	Qwen3.8-Flash-NextIQ3_XXS	RTX 5070 12GB	66.7	—	Strata	@jun1228909
30NEW	AliceAI-Foundation-80B-A3BQ4_K_M	M5 Max	65	—	llama.cpp	@aqty
31NEW	Qwen3.8-Flash-Next4-bit QAD	DGX Spark	65	—	TensorFold	@sudoingX
32NEW	Qwen3.8-27B4-bit	M6 mini 32GB	61.9	—	TensorFold	@jmurillocode
33NEW	GLM 5.3 FlashEXL3	DGX Spark	60.4	1950	TensorFold	@MiaAI_lab
34NEW	Qwen3.8-Flash-NextFP8	DGX Spark	58	—	vLLM	@sudoingX
35NEW	DeepSeek v4.1 FlashQ2	RTX PRO 6000	57.7	—	llama.cpp b11412	@piyo621
36NEW	GLM 5.3 FlashNVFP4	DGX Spark	55.6	—	—	@PlusTen_AI
37NEW	Qwen3.8-Flash-NextUD-IQ4_XS	Ryzen AI Max+ 395 128GB	53.8	1293	Strata 0.1.40	@Niko1221
38NEW	Qwen3.8-Flash-NextIQ3_S	RTX 5070 Ti 16GB	53.5	—	Strata	@raminaol9ykas
39NEW	Qwen3.8-Flash-NextIQ3_S	RTX 3080 10GB	52.7	—	Strata	@needmorevram
40NEW	Qwen3.8-Flash-NextIQ3_S	2× AMD MI50 32GB	52	645	Strata	@liucl
41NEW	Qwen3.8-Flash-Next CoderGSQ-RCO	M5 Max	51	—	llama.cpp	@sebastavar
42NEW	Qwen3.8-Flash-Next4bit	DGX Spark	50.2	—	Vontra MLX	@openaiarka
43NEW	Empero Qwen3.8-35B-A3BQ4_K_M	RTX 3060	50	550	llama.cpp	@Oluwaphilemon1
44NEW	Qwen3.8-Flash-NextIQ3_S	RTX 4060 Ti	50	—	Strata	@xiaobei1943
45NEW	Qwen3.5-9B4bit	M4 Pro	49.3	—	rapid-mlx	@rapidmlx
46▼42	Qwen3-8B4bit	M4 Pro	48.3	—	rapid-mlx	@rapidmlx
47NEW	MiMo-V2.6-Distill-Qwen-9BQ5_K_M	RTX 5060	47	1600	llama.cpp	@stfu0911
48NEW	Qwen3.8-Flash-NextFP8	DGX Spark	45	—	—	@Oluwaphilemon1
49NEW	Qwen3.8-27BNVFP4	DGX Spark	42.3	—	TensorFold 0.6.3	@edurdias
50NEW	Qwen3.8-27BUD-Q4_K_XL	RTX 4090	40.7	2659.8	llama.cpp	@Oluwaphilemon1
51NEW	Ornith	GTX 1660 SUPER	40	—	—	@mine_craft_bui
52NEW	Qwen3.8-Flash-NextIQ3_S	2× RTX 3060 12GB	40	—	Strata	@xr_ymchn
53NEW	GLM 5.3EXL3 3.0bpw	DGX Spark	38.4	—	TensorFold	@niklaslenz_ai
54▼49	Qwen3.6-35B-A3BUD-IQ3_XXS	M1 Max 32GB	37	350	—	@moonsteroid
55NEW	GLM 5.3NVFP4	DGX Spark	35	—	TensorFold	@majewskizby
56NEW	GLM 5.3 FlashEXL3 2.05bpw	DGX Spark	32.3	—	TensorFold	@NFT_Chen
57NEW	Qwen3.8-Flash-NextNVFP4	DGX Spark	32	—	TensorFold	@openaiarka
58NEW	Qwen3.8-27BNVFP4	DGX Spark	31.8	—	SGLang 0.5.21	@edurdias
59NEW	GLM 5.3Int4/Int8	DGX Spark	30	840	sparkDash	@majewskizby
60NEW	Qwen3.8-27BNVFP4	DGX Spark	28.7	—	vLLM 0.30.0	@edurdias
61NEW	Qwen3.8-Flash-NextIQ2_XS	RTX 3060	27.7	—	Strata	@dec21ai
62NEW	Qwen3.8-Flash-NextIQ3_S	AMD MI50 32GB	26.8	—	Strata	@yokoyoko001
63NEW	Qwen3.8-27BIQ3_XXS	RTX 5060 Ti	23	—	llama.cpp	@fntAInhead
64NEW	Qwen3.8-27BUD-Q2_K_XL	RTX 5060 Ti	19	—	llama.cpp	@fntAInhead
65NEW	GLM 5.3 FlashEXL3	DGX Spark	18.6	—	EXL3	@vcruz305
66NEW	GLM 5.3 FlashNVFP4	DGX Spark	15	—	vLLM	@sudoingX
67NEW	Nex-N2.5-miniQ4_K_M	RTX 5060 Ti	14	—	llama.cpp	@fntAInhead
68NEW	Qwen3.8-27B TurboFCFusionUncensoredIQ2_M	RTX 5060 Ti	10	—	llama.cpp	@fntAInhead
69NEW	Qwen3.8-Flash-NextQ4_K_M	M5 Max	7.8	—	LM Studio	@miya_mos365

GLM 5.3

DGX Spark

4× DGX Spark · TensorFold · NVFP4 + FP4 KV

Pair

#55 / 69 in this filter

#15 / 22 on this chip

#2 / 3 of this model

Decode · tok/s

Prefill

Peak

On this chip

This model elsewhere

Lineage

Officialclaimed

Originating sentence

Full GLM-5.3 (753B) on 4x DGX Spark. Context 64K → 256K. Sparkdash: prose c1 32.7 → 35.0 tok/s, code 37.9 → 42.0. FP4 KV + NVFP4 weights for attention/shared/MTP.

Flags

Sparkdash c1 · 256K · FP4 KV · NVFP4 on attention/shared/MTP

Caveat

First-person 4× Spark update. Rank is the quoted Sparkdash prose 35.0. Code 42.0 stays in the note. Separate from the Int4/Int8 30.0 prose pair and the EXL3 3.0bpw 38.4 code pair.

Proof

@majewskizby

Get

Get unlinked

Used boxes

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