Which model leads C-Eval?

Q: Which model leads C-Eval?

Qwen3 Max (Reasoning) tops the board at 93.7

C-Eval Leaderboard — 49 Models Ranked

Pro Mode — raw scores, one benchmark under the lensofficial card harness

The field · 49 measured

40.793.7

49 models measured, most on the official card harness. Qwen3 Max (Reasoning) tops the board at 93.7.

49 measured·Lifecycle Updated Oct 8, 2026Score basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 49Bars show 40.7–93.7 · ◆ solid = first-party or better · outlined = arm's-length

01Qwen3 Max (Reasoning)Alibaba93.7$1.2/M$6.0/M$0.038

02Qwen3.6 PlusAlibaba93.3$0.50/M$3.0/M$0.019

03DeepSeek-V4-Pro-Base+2 altsDeepSeek93.1———

04Qwen3.5 397B A17B+1 altAlibaba93$0.60/M$3.6/M$0.023

05Kimi K2 Base+8 altsMoonshot92.5———

06Claude Opus 4.5Anthropic92.23P$5.0/M$25.0/M$0.163

07DeepSeek-V4-Flash-Base+2 altsDeepSeek92.1———

08Hunyuan LargeTencent91.9———

09Qwen3.5 122B A10BAlibaba91.9$0.40/M$3.2/M$0.020

10DeepSeek-R1DeepSeek91.8$2.0/M$4.0/M$0.033

11MiMo V2.5 Pro Base+2 altsXiaomi91.5———

12MiMo V2.5 ProXiaomi91.5$0.43/M$0.87/M$0.007

13Qwen3.6 27B+1 altAlibaba91.4$0.60/M$3.6/M$0.023

14DeepSeek-V3.2-Exp-Base+2 altsDeepSeek913P———

15Qwen2 72BAlibaba91———

16Qwen3.5 27B+2 altsAlibaba90.5$0.30/M$2.4/M$0.015

17DeepSeek-V3-Base+1 altDeepSeek90.35———

18Qwen3.5 35B A3B+1 altAlibaba90.2$0.25/M$2.0/M$0.012

19Qwen3.6 35B A3B+2 altsAlibaba90$0.38/M$2.3/M$0.015

20DeepSeek-V3.1-Base+2 altsDeepSeek903P———

21Hy3-preview-Base+1 altTencent89.8———

22Step3.5 Flash Base+1 altStepFun89.63P———

23Qwen2.5 Instruct 72BAlibaba89.2$0.47/M$0.49/M$0.005

24MiMo V2.5 Base+2 altsXiaomi88.6———

25Hunyuan Large InstructTencent88.6———

26Qwen3.5 9BAlibaba88.2$0.14/M$0.20/M$0.002

27MiMo V2 Flash Base+2 altsXiaomi87.93P———

28DeepSeek-V3+1 altDeepSeek86.5$0.24/M$0.90/M$0.007

29Qwen3 14BAlibaba86.2$0.35/M$1.4/M$0.010

30GLM-4.5-Base+3 altsZ.ai85.84———

31Qwen3.5 4BAlibaba85.1$0.03/M$0.15/M$0.001

32Qwen2 Instruct (72B)Alibaba83.8———

33Gemma 4 31B+1 altGoogle82.63P$0.17/M$0.40/M$0.003

34Gemma 4 26B A4BGoogle82.53P$0.07/M$0.34/M$0.002

35DeepSeek-V2+2 altsDeepSeek81.4———

36Moonlight+1 altMoonshot77.23P———

37Qwen2 7B InstructAlibaba77.2———

38Claude 3.5 SonnetAnthropic76.7$3.0/M$15.0/M$0.117

39GPT-4oOpenAI76$5.0/M$15.0/M$0.132

40Qwen2.5 3B+1 altAlibaba753P———

41Qwen3.5 2BAlibaba73.2———

42Llama 3.1 Instruct 405BMeta72.5———

43O1 MiniOpenAI68.9———

44Llama 3 (70B)Meta65.2———

45Qwen3.5 0.8BAlibaba50.5———

46Llama 3 Instruct 8BMeta49.5$0.04/M$0.14/M$0.002

47Mistral 7BMistral47.4———

48Gemma (7B)Google43.6———

49ERNIE 4.5Baidu40.7———

◆ solid = independently verified · aggregator attested · vendor attributed — outlined = cross-reference · source-attributed. Dimmed ◆ = confidence inferred. “+N alts” = alternate disclosures in the drawer with why-demoted. Prices per million tokens; $/point = input price ÷ score. Every row opens its model dossier.

Verification: 49 models scored · 0 independently verified · 19 vendor cross-reference · 30 vendor-reported · 0 with source disagreement. How these tiers are assigned