Skip to content
VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Models741Deals30d269Policy30d65, 14 pendingSources1,014

LongBench v2 leaderboard — the family board.

14 TRACKED VARIANTS · 1 FIRST-CLASS BOARD

One family, 14 tracked prints. Scoring configs and splits of the base board render as sections of this page, each under its own honest label.

● LIVE DATA38 MODELS ON THIS BOARDLEADER CLAUDE OPUS 4.8 AT 69.1

The board — LongBench v2TOP 38 OF 38 MEASURED · BEST SCORE PER MODEL, DEFAULT VARIANT

#ModelVendorScoreVLast seen
01Claude Opus 4.8Anthropic69.12026-08-24
02Qwen3.5 397B A17BAlibaba68.92026-08-23
03Gemini 3 ProGoogle68.22026-06-15
04GPT-5.6 SolOpenAI67.12026-08-24
05Qwen3.8 Max PreviewAlibaba66.32026-08-24
06Qwen3.7 MaxAlibaba65.32026-08-24
07Gemini 2.5 ProGoogle652026-06-05
08Claude Opus 4.5Anthropic64.42026-06-15
09Qwen3.6 PlusAlibaba622026-08-23
10Nemotron 3 Ultra 550B A55BNVIDIA61.92026-09-01
11Claude Sonnet 4.5Anthropic61.82026-06-06
12MiniMax M1 80KMiniMax61.52026-08-23
13Kimi K2.5Moonshot612026-08-23
14MAI-Thinking-1Microsoft612026-08-23
15MiniMax M1 40KMiniMax612026-08-23
16Qwen3 Max (Reasoning)Alibaba60.62026-09-09
17MiMo V2 FlashXiaomi60.62026-08-23
18Qwen3.5 27BAlibaba60.62026-08-23
19Qwen3.5 122B A10BAlibaba60.22026-08-23
20DeepSeek-V3.2DeepSeek59.82026-06-15
21Qwen3.5 35B A3BAlibaba592026-08-23
22o3OpenAI58.82026-06-05
23DeepSeek-R1DeepSeek58.32026-06-05
24Claude Opus 4Anthropic55.62026-06-05
25Qwen3.5 9BAlibaba55.22026-08-23
26GPT-5.2OpenAI54.52026-06-15
27Qwen3 235B A22BAlibaba50.12026-06-05
28Qwen3.5 4BAlibaba502026-08-23
29DeepSeek-V3DeepSeek48.72026-08-23
30GPT-4oOpenAI48.12026-05-03
31Kimi K2 ThinkingMoonshot45.12026-06-13
32Claude 3.5 SonnetAnthropic412026-05-03
33Qwen2.5 Instruct 72BAlibaba39.42026-05-03
34Qwen3.5 2BAlibaba38.72026-08-23
35Llama 3.1 Instruct 405BMeta36.12026-05-03
36DeepSeek-V2.5DeepSeek35.42026-05-03
37DeepSeek-V2DeepSeek31.62026-05-03
38Qwen3.5 0.8BAlibaba26.12026-08-23
◆ IV — INDEPENDENTLY VERIFIED · ◆ AA — AGGREGATOR · ◆ VA — VENDOR · ◇ SA — SOURCE-ATTRIBUTED

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

LongBench v2 overall (w/o CoT)

VARIANT5 models · best score per model
#ModelVendorScoreVLast seen
01MiniMax Text 01MiniMax52.92026-06-06
02GPT-4oOpenAI50.12026-06-06
03DeepSeek-V3DeepSeek48.72026-06-06
04Qwen2.5 Instruct 72BAlibaba42.12026-06-06
05Claude 3.5 SonnetAnthropic412026-06-06

Other tracked prints

THIN BOARDSunder 5 published models each — listed, never hidden
PrintLeaderScoreModels
LongBench v2 easy (w/ CoT)MiniMax Text 0166.14
LongBench v2 easy (w/o CoT)MiniMax Text 0160.94
LongBench v2 hard (w/ CoT)MiniMax Text 0150.54
LongBench v2 hard (w/o CoT)MiniMax Text 0147.94
LongBench v2 long (w/ CoT)MiniMax Text 0147.24
LongBench v2 long (w/o CoT)Qwen2.5 Instruct 72B44.44
LongBench v2 medium (w/ CoT)MiniMax Text 0156.74
LongBench v2 medium (w/o CoT)MiniMax Text 0152.64
LongBench v2 overall (w/ CoT)MiniMax Text 0156.54
LongBench v2 short (w/ CoT)MiniMax Text 0161.74
LongBench v2 short (w/o CoT)MiniMax Text 0158.94
LongBench-V2 (EM)DeepSeek-V4-Pro-Base51.53
THE FAMILY BOARD · EVERY PRINT LABELED · EVERY SCORE SOURCE-ATTRIBUTED ON ITS OWN LIVE BOARDAll leaderboardsThe frontier boardModels by vendor