Skip to content
VECTOR WIREAI INTELLIGENCE
UTC

MGSM leaderboard — the family board.

3 TRACKED VARIANTS · 1 FIRST-CLASS BOARD

One family, 3 tracked prints. Scoring configs and splits of the base board render as sections of this page, each under its own honest label.

● LIVE DATA56 MODELS ON THIS BOARDLEADER LLAMA 4 MAVERICK AT 92.3

The board — MGSMTOP 56 OF 56 MEASURED · BEST SCORE PER MODEL, DEFAULT VARIANT

#ModelVendorScoreVLast seen
01Llama 4 MaverickMeta92.32026-09-11
02o3-miniOpenAI922026-08-23
03Llama 3.1 Instruct 405BMeta91.62026-09-11
04Claude 3.5 SonnetAnthropic91.62026-08-23
05Llama 3.3 70B InstructMeta91.12026-09-11
06o1OpenAI90.82026-08-23
07Claude 3 OpusAnthropic90.72026-08-23
08Llama 4 ScoutMeta90.62026-08-23
09GPT-4oOpenAI90.52026-08-23
10GPT-4 TurboOpenAI88.52026-08-23
11Nemotron 3 Ultra 550B A55B BaseNVIDIA87.72026-06-05
12Gemini 1.5 ProGoogle87.52026-08-23
13Qwen2.5 Instruct 72BAlibaba87.32026-05-30
14Gemma 3 4BGoogle87.32026-10-07
15GPT-4o miniOpenAI872026-08-23
16Llama 3.1 70B InstructMeta86.92026-05-01
17Claude 3.5 HaikuAnthropic85.62026-08-23
18Kimi K2 BaseMoonshot85.22026-06-05
19Qwen3 235B A22BAlibaba83.52026-08-23
20Claude 3 SonnetAnthropic83.52026-08-23
21Mistral Large 3 675B Base 2512Mistral82.92026-06-05
22DeepSeek-V3.2-Exp-BaseDeepSeek82.32026-06-05
23Granite 4.1 8BIBM82.32026-06-15
24Qwen3 4B 2507 InstructAlibaba81.82026-10-07
25GLM-4.5-BaseZ.ai81.32026-06-05
26Phi 4Microsoft80.62026-08-23
27DeepSeek-V3DeepSeek79.82026-05-01
28Qwen 2.5 14BAlibaba79.62026-05-30
29Claude 3 HaikuAnthropic75.12026-08-23
30GPT-4OpenAI74.52026-08-23
31Granite 4.1 30B BaseIBM74.42026-06-15
32LFM2-2.6BLiquid AI74.32026-10-07
33Granite 4.0 H TinyIBM73.72026-10-07
34LFM2-8B-A1B (Non-Reasoning)Liquid AI72.42026-10-07
35Granite 4.1 30BIBM71.12026-06-15
36Granite 4.1 3BIBM702026-06-15
37Llama 3.1 8B InstructMeta68.92026-05-01
38Qwen3 1.7BAlibaba66.62026-10-07
39Phi 4 Mini InstructMicrosoft63.92026-08-23
40Granite 4.1 8B BaseIBM63.72026-06-15
41DeepSeek-V2DeepSeek63.62026-05-01
42Llama 3.2 3B InstructMeta61.72026-10-07
43Phi 3.5 MoE InstructMicrosoft58.72026-08-23
44Granite 4.1 3B BaseIBM58.52026-06-15
45LFM2-1.2BLiquid AI55.02026-10-07
46Phi 3 Medium 4K InstructMicrosoft53.52026-05-30
47Phi 3.5 Mini InstructMicrosoft47.92026-08-23
48LFM2-700MLiquid AI45.42026-10-07
49Gemma 3 1B InstructGoogle43.62026-10-07
50Qwen3 0.6BAlibaba41.32026-10-07
51Qwen2.5 7B BaseAlibaba36.52026-05-03
52Qwen2.5 3B BaseAlibaba342026-05-03
53Qwen3 4B Base DenseAlibaba31.02026-05-03
54LFM2-350MLiquid AI29.52026-10-07
55Llama 3.2 Instruct 1BMeta29.12026-10-07
56Gemma 3 1B Base DenseGoogle1.72026-05-03
◆ IV — INDEPENDENTLY VERIFIED · ◆ AA — AGGREGATOR · ◆ VA — VENDOR · ◇ SA — SOURCE-ATTRIBUTED

Variants of this boardSCORING CONFIGS & SPLITS — NOT CITED STANDALONE, SO THEY LIVE HERE

MGSM (EM)

VARIANT7 models · best score per model
#ModelVendorScoreVLast seen
01DeepSeek-V4-Flash-BaseDeepSeek85.72026-09-10
02DeepSeek-V4-Pro-BaseDeepSeek84.42026-09-10
03DeepSeek-V3.2-BaseDeepSeek81.32026-06-27
04DeepSeek-V3DeepSeek79.82026-07-06
05Qwen2.5 Instruct 72BAlibaba76.22026-07-06
06Llama 3.1 Instruct 405BMeta69.92026-07-06
07DeepSeek-V2DeepSeek63.62026-07-06

Other tracked prints

THIN BOARDSunder 5 published models each — listed, never hidden
PrintLeaderScoreModels
MGSM-rev2Qwen3.5 2B86.52
THE FAMILY BOARD · EVERY PRINT LABELED · EVERY SCORE SOURCE-ATTRIBUTED ON ITS OWN LIVE BOARDAll leaderboardsThe frontier boardModels by vendor