Which model leads TruthfulQA?

Q: Which model leads TruthfulQA?

MAI-Thinking-1 tops the board at 88

TruthfulQA Leaderboard — 26 Models Ranked

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

The field · 26 measured

42.288

26 models measured, most on the official card harness. MAI-Thinking-1 tops the board at 88.

26 measured·Lifecycle Updated Oct 8, 2026Score basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 26Bars show 42.2–88 · ◆ solid = first-party or better · outlined = arm's-length

01MAI-Thinking-1Microsoft88———

02Phi 3.5 MoE InstructMicrosoft77.5———

03Granite 3.2 8B InstructIBM66.923P———

04Granite 3.3 8B Instruct+1 altIBM66.86$0.03/M$0.25/M$0.002

05Phi 4 Mini InstructMicrosoft66.4———

06Granite 3.1 8B InstructIBM65.843P———

07Phi 3 Mini 4K InstructMicrosoft64.73P———

08Phi 3.5 Mini InstructMicrosoft64———

09Llama 3 Instruct 8B+1 altMeta63.2$0.04/M$0.14/M$0.002

10Granite 3.2 2B InstructIBM59.83P———

11Granite 3.1 2B InstructIBM59.793P———

12Granite 3.3 2B InstructIBM58.973P———

13Llama 3.1 Nemotron Instruct 70BNVIDIA58.63———

14Jamba 1.5 LargeAI2158.3$2.0/M$8.0/M$0.086

15Command R+ (Apr '24)Cohere56.3$2.5/M$10.0/M$0.111

16Qwen2 Instruct (72B)Alibaba54.8———

17Qwen2 72BAlibaba54.8———

18Qwen 2.5 Coder 32B InstructAlibaba54.2$0.06/M$0.20/M$0.002

19Jamba 1.5 MiniAI2154.1$0.20/M$0.40/M$0.006

20Llama 3.1 8B InstructMeta52.793P$0.02/M$0.04/M$0.001

21DeepSeek-R1-Distill-Llama-8BDeepSeek47.433P$0.05/M$0.05/M$0.001

22DeepSeek-R1-Distill-Qwen-7BDeepSeek47.143P$0.15/M$0.15/M$0.003

23Llama 3 (70B)Meta45.6———

24Gemma (7B)+1 altGoogle44.8———

25DeepSeek-V2DeepSeek42.2———

26Mistral 7B+1 altMistral42.2———

◆ 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: 26 models scored · 0 independently verified · 13 vendor cross-reference · 13 vendor-reported · 0 with source disagreement. How these tiers are assigned