Which model leads Arena-Hard (Hard Prompt)?

Gemini 3 Pro holds the best tracked Arena-Hard (Hard Prompt) score (72.6) across 6 measured models on Vector Wire.

SEEDRefresh Models —Deals —Regulatory —Sources —

Arena-Hard (Hard Prompt)

6 models tracked

Arena-Hard-Auto (Hard Prompt subset) — Explicit category subset of Arena-Hard, distinct domain from Creative Writing subset

#	Model	Vendor	Best score	Runs	Last seen
1	Gemini 3 Pro	Google	72.6	2	2026-06-13
2	GPT-5	OpenAI	71.9	1	2026-06-06
3	Kimi K2 (Reasoning)	Moonshot	71.9	2	2026-06-13
4	Claude Sonnet 4.5	Anthropic	63.3	1	2026-06-06
5	MiMo V2 Flash	Xiaomi	54.1	2	2026-06-13
6	DeepSeek-V3.2	DeepSeek	53.4	2	2026-06-13

Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.