[mle]/DaBench

Independent, real-world evaluation of AI models

Open-weight and hosted models take the same private test suite, graded by code instead of another model. Every score links to the run behind it.

What they build

See all 94 builds
1Claude Haiku 5.570Showcase225 tok/s$0.42 per run
2Muse Spark 1.3 Contributor66107 tok/s$0.41 per run
3DeepSeek V4.1 Flash64Open weights30 tok/s$0.39 per run
4GPT-6 Luna61Showcase90 tok/s$0.10 per run
5Beam 1.0 501B A23B57Open weights274 tok/s$0.83 per run
6DeepSeek V4.1 Flash 101756Open weightsShowcase243 tok/s$0.75 per run
7Qwen3.8 27B48Open weights196 tok/s$1.28 per run
7MiMo-V2.6-Pro48Open weights55 tok/s$1.80 per run
8Ling 3.1 Flash45Open weightsShowcase112 tok/s
8MiMo-V2.6-Flash45Open weightsShowcase66 tok/s$0.33 per run
9Qwen3.8 Flash Next42Open weights68 tok/s$0.46 per run
10MiniMax M3.1 Flash35Open weightsShowcase135 tok/s
10GLM 5.3 Flash35Open weights60 tok/s$0.19 per run
11Underdog Saluki 27B33Open weightsShowcaseIQ2_XS10.2 GB at 32K23 tok/s on M5 Pro
12Clover 35B A3B Preview30Open weightsShowcase252 tok/s
13Mistral Large 428Showcase56 tok/s$2.67 per run
14Qwen3.6 35B A3B26Open weights156 tok/s$1.47 per run
15Apodex 1.1 Mini (free)25Open weights278 tok/s
16Flint21Open weightsShowcase149 tok/s
17Spark X2.6 4B15Open weightsShowcaseBF169.4 GB at 32K136 tok/s on RTX 5090$1.61 per runest.
18Mellum2.1 Thinking9Open weightsQ8_012.9 GB at 32K90 tok/s on M5 Pro$2.46 per runest.
19LFM2.5 24B A2B7Open weightsQ8_025.0 GB at 32K82 tok/s on M5 Pro
20LFM3 1.7B6Open weightsQ8_02.8 GB at 32K104 tok/s on M5 Pro
21Spark X2.6 1.7B4Open weightsQ8_02.9 GB at 32K106 tok/s on M5 Pro

Each model ranks by its best build. Equal overall scores share a rank. How scores work

Compare two models

One email when a new model lands on the board or in the showcase. Unsubscribe anytime.