Which model leads Arena-Hard (Hard Prompt)?

Q: Which model leads Arena-Hard (Hard Prompt)?

Gemini 3 Pro tops the board at 72.6

Arena-Hard (Hard Prompt) Leaderboard

Pro Mode — raw scores, one benchmark under the lensthird-party eval harness

The field · 6 measured

53.472.6

6 models measured, most on the third-party eval harness. Gemini 3 Pro tops the board at 72.6.

6 measured·Lifecycle Updated Oct 8, 2026Score basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 6Bars show 53.4–72.6 · ◆ solid = first-party or better · outlined = arm's-length

01Gemini 3 ProGoogle72.6$2.0/M$12.0/M$0.096

02GPT-5OpenAI71.9$1.3/M$10.0/M$0.078

03Kimi K2 ThinkingMoonshot71.9$0.60/M$2.5/M$0.022

04Claude Sonnet 4.5Anthropic63.3$3.0/M$15.0/M$0.142

05MiMo V2 FlashXiaomi54.1$0.10/M$0.30/M$0.004

06DeepSeek-V3.2DeepSeek53.4$0.28/M$0.42/M$0.007

◆ 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: 6 models scored · 0 independently verified · 5 vendor cross-reference · 1 vendor-reported · 0 with source disagreement. How these tiers are assigned