Ring-1T Report - ArenaHard v2 (win-rate): leaderboard
ArenaHard v2: difficult user prompts whose answers an LLM judge compares with a baseline model's answers; win rate in percent. From Table 1 of inclusionAI's Ring-1T technical report, comparing Ring-1T with DeepSeek-V3.1-Terminus (Thinking), Qwen3-235B-A22B-Thinking-2507, Gemini 2.5 Pro and GPT-5 Thinking (High), evaluated with a 32K context window. Baselines use official hyperparameters for open models and vendor-recommended settings for APIs, and where an official score exists the report keeps the higher of its reproduction and the official figure.
Metric: ArenaHard v2 (win-rate). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (High) | 82.91 | #163 (GPT-5) |
| 2 | Ring-1T | 81.59 | #628 |
| 3 | Qwen 3 235B A22B 2507 (Thinking) | 80.18 | #460 |
| 4 | Gemini 2.5 Pro | 79.2 | #262 |
| 5 | DeepSeek V3.1 Terminus (Thinking) | 60.27 | #442 (DeepSeek V3.1 Terminus) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ring-1t-report-arenahard-v2-win-rate · How It Works · Data refreshed daily, snapshot 2026-11-12.