Which model leads MT-Bench?

Q: Which model leads MT-Bench?

Hunyuan Large Instruct tops the board at 9.4

MT-Bench Leaderboard — 16 Models Ranked

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

The field · 16 measured

0.099.4

16 models measured, most on the official card harness. Hunyuan Large Instruct tops the board at 9.4.

16 measured·Lifecycle Updated Oct 9, 2026Score basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 16Bars show 0.09–9.4 · ◆ solid = first-party or better · outlined = arm's-length

01Hunyuan Large InstructTencent9.4———

02Qwen2.5 Instruct 72BAlibaba9.35$0.47/M$0.49/M$0.052

03DeepSeek-V2.5+1 altDeepSeek9.02———

04Claude 3.5 SonnetAnthropic8.81$3.0/M$15.0/M$1.022

05GPT-4oOpenAI8.74$5.0/M$15.0/M$1.144

06Mistral Large 2Mistral8.63$2.0/M$6.0/M$0.463

07LFM2.5-8B-A1B-DSpark+1 altLiquid AI8.52———

08LFM2.5-8B-A1BLiquid AI8.52———

09Llama 3.1 Instruct 405B+1 altMeta8.49———

10Qwen2 7B InstructAlibaba8.41———

11Llama 3.1 70B Instruct+1 altMeta8.22$0.56/M$0.56/M$0.068

12Llama 3.1 Nemotron Nano 8B V1NVIDIA8.1———

13LFM2.5-2.6B-DSpark+1 altLiquid AI5.07———

14LFM2.5-2.6BLiquid AI5.07———

15LFM2.5-1.2B-Instruct+1 altLiquid AI3.9———

16Llama 3.1 Nemotron Instruct 70B+1 altNVIDIA0.09———

◆ 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: 16 models scored · 0 independently verified · 4 vendor cross-reference · 12 vendor-reported · 2 with source disagreement. How these tiers are assigned