localairecipes launch configs for local inference

Leaderboard

Refine

Source:

Model
Quant
Storage requirement
Feature tags

NVIDIA DGX Spark 128GB ×1 (Blackwell)

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
GLM-5.3-Flash VQ Q4_K 1x DGX Spark 1M ctx (JLFN) #14224.3 tok/s12.1 tok/s———reported0
Qwen3.8-Flash-Next NVFP4 DGX Spark GB10 SGLang NEXTN MTP40110.4 tok/s———465 msreported0
Qwen3.8-Flash-Next W4A16-AutoRound DGX Spark vLLM MTP4 Iter7g4076.8 tok/s179.9 tok/s261.2 tok/s——reported0
Qwen3.8-Flash-Next NVFP4 1x DGX Spark vLLM 71.3 c1 (salient-data)4069.3 tok/s—91.9 tok/s——reported0
Qwen3.8-Flash-Next FP4-Local 1x DGX Spark vLLM b12x4066.2 tok/s166.4 tok/s——42.2 msreported0
Qwen3.8-Flash-Next Uncensored NVFP4 BF16 PLE DGX Spark vLLM MTP24044.2 tok/s——2055.2 tok/s116800 msreported0
Qwen3.8-Flash-Next NVFP4 DGX Spark GB10 SGLang MTP-213 262K4032.7 tok/s93.2 tok/s——286 msreported0
Qwen3.8-27B SGLang NVFP4 DFlash2 on 1x DGX Spark (deterministic)3468.3 tok/s81.7 tok/s141.5 tok/s——reported0
Nemotron 3.5 Bolt 74B-A9B 1x DGX Spark vLLM b12x LiLiCorr 262Kn/a140 tok/s113.4 tok/s163.6 tok/s——reported0
GLM-5.3-Flash exlr8 EXL3 experts on DGX Spark TP1-4n/a128 tok/s——4393 tok/s—reported0

NVIDIA RTX PRO 6000 Blackwell 96GB ×2

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
GLM-5.3-Flash QAD (CSF) on 2x RTX PRO 6000 Blackwell — TP2/DCP2 MTP3 #142207 tok/s488 tok/s672 tok/s——reported0
DeepSeek V4 Flash Vision on 2x RTX PRO 6000 Blackwell \x{2014} TP2 DSpark K335169.2 tok/s394.7 tok/s—8993 tok/s—reported0
DeepSeek V4 Flash-0731 on 2x RTX PRO 6000 Blackwell \x{2014} TP2 DSpark K534214.9 tok/s—611.9 tok/s11167 tok/s—reported0

NVIDIA CMP 170HX 64GB ×6 (Ampere)

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
DeepSeek-V4.1-Flash SM80 DeepInfer kernel port 6x CMP 170HX DSpark K5 #139160.8 tok/s170.5 tok/s—4506.8 tok/s10700 msreported0

Machine-readable: /api/v1/leaderboard.json (these boards) · /api/v1/scores.json (per-recipe medians).