Leaderboard Results at a Glance
The early May 2026 LM Arena text model leaderboard update reveals a noteworthy trend: the collective rise of Chinese models in the global arena.
Key Rankings
| Global Rank | Model | Company | Country | Change |
|---|---|---|---|---|
| 13 | Ernie 5.1 Preview | Baidu | 🇨🇳 | Stable |
| 16 | GPT-5.5 | OpenAI | 🇺🇸 | ↓ |
| 22 | mimo-v2.5-pro | Xiaomi | 🇨🇳 | ↑ |
| 23 | DeepSeek V4 Pro | DeepSeek | 🇨🇳 | ↑ |
Data Interpretation
Ernie 5.1 Preview holds Chinese #1: At global rank 13, Baidu’s Ernie 5.1 Preview remains the highest-ranked Chinese model — stable in the top 15% globally.
GPT-5.5 no longer “solidly holding”: Rank 16 represents a decline. Evaluators note it “no longer solidly catches” — suggesting OpenAI’s flagship model’s advantage is narrowing as Chinese models catch up.
Xiaomi and DeepSeek’s upward momentum: Both models surpassed GPT-5.5’s position, unthinkable just six months ago.
Landscape Judgment
Chinese Models: From “Catching Up” to “Running Alongside”
Current Chinese model distribution in LM Arena:
Global 10-15: Ernie 5.1 Preview (Chinese ceiling)
Global 20-25: mimo-v2.5-pro, DeepSeek V4 Pro (catching-up group)
Global 25-35: GLM-5.1, Kimi K2.6, Qwen 3.6 (main force)Chinese models have collectively moved up 5-10 positions compared to six months ago.
US Models: Advantage is Shrinking
GPT-5.5’s decline signals:
- Chinese models’ improvement speed exceeds US models’ iteration speed
- Price factors begin influencing voting behavior
- “US models only” is no longer the optimal strategy
Actionable Recommendations
| Decision Dimension | US Models | Chinese Models |
|---|---|---|
| Absolute Performance | Still leading | Gap rapidly closing |
| Price | Higher | Significantly lower |
| Compliance Risk | Exists | None |
| Local Support | Limited | Comprehensive |
For China-market businesses, Chinese models’ advantages in compliance, cost, and localization can already compensate for minor performance gaps.