localairecipes launch configs for local inference

Leaderboard

Refine

Source:

Model
Quant
Storage requirement
Feature tags

NVIDIA DGX Spark 128GB ×3 (Blackwell)

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
GLM-5.3-Flash exlr8 EXL3 experts on DGX Spark TP1-4 #1n/a128 tok/s——4393 tok/s—reported0

NVIDIA RTX PRO 6000 Blackwell 96GB ×4

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
GLM-5.3-Flash on 4x RTX PRO 6000 Blackwell \x{2014} TP4 MTP3 #142248.9 tok/s—873.8 tok/s13514 tok/s—reported0
GLM-5.3-Flash on 4x RTX PRO 6000 Blackwell \x{2014} TP4 DFlash2 K742219.7 tok/s—677.7 tok/s13840 tok/s—reported0
DeepSeek V4.1 Flash on 4x RTX PRO 6000 Blackwell \x{2014} TP4 DSpark K739230.6 tok/s—748.3 tok/s17503 tok/s—reported0

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