Top AI Models of July 2025: Discover the 10 Best Performers!

With AI agents becoming increasingly popular, tech companies are pushing the boundaries of innovation to introduce more advanced models. LMArena aims to keep pace with this rapid development by offering dynamic rankings of the most efficient models. Here are the major trends in July 2025.

Top AI models of July 2025

Has Gemini permanently secured the top position in the LMArena rankings? As of July 2025, the Gemini-2.5-Pro model has been more dominant than ever in the overall rankings (see featured image). It leads in every category reviewed. The rest of the podium includes two models from OpenAI: o3 and ChatGPT-4o, which, however, are significantly outperformed by the Gemini-2.5-Pro.

The rest of the rankings highlight impressive performances by Grok (4th), which has not been affected by its recent setbacks, and Kimi (6th), developed by the Chinese startup Moonshot AI. Claude, having been off the radar for several months, continues to benefit from LMArena’s new ranking system, managing to place two of its models in the top 10 by July 2025.

Here are the top 10 AI models in July 2025, according to LMArena:

1. Gemini-2.5 Pro
2. OpenAI o3
3. ChatGPT 4o
4. Grok-4
5. GPT-4.5
6. Kimi-k2
7. Claude Opus 4 (“thinking” version)
8. Claude Opus 4 (standard version)
9. DeepSeek R1
10. Gemini-2.5 Flash

LMArena: Category-specific rankings

Since last June, LMArena has been offering detailed rankings for different fields. Here are the trends for July 2025:

• Text: Gemini-2.5-Pro remains slightly ahead, closely followed by OpenAI. The competition is tight, but the balance at the top shows a refined equilibrium among rivals.
• Web Development: Gemini, DeepSeek, and Claude, in various versions, lead this category. OpenAI only makes it to 7th place here.
• Image Analysis: Gemini confirms its lead, but the margins are narrowing. It is closely followed by OpenAI models.
• Online Search: The top spots are shared between Gemini and Perplexity’s Sonar models, which are prevalent in this category. OpenAI remains less visible, relegated to background roles with its API variants.
• Code Assistants: DeepSeek takes the top spot, but Claude’s 3.5 Sonnet models are close contenders. Mistral makes its first significant appearance at the top of the rankings.
• Image Generation: GPT-Image-1 remains at the forefront despite stiff competition from Flux. The challengers organize behind, yet still fall short of the leader’s scores.

What are LMArena’s ranking criteria?

Developed by the Large Model Systems Organization (LMSYS), LMArena uses a system of anonymous duels to assess the performance of AI models. In each instance, two models are blindly compared, and users vote for the one that provided the best response. Each outcome affects an Elo score, inspired by the chess ranking system: defeating a higher-ranked model boosts the score, while losing to a lower-ranked one lowers it.

This mechanism allows for the generation of several distinct rankings based on the tasks or fields evaluated. From these specific rankings, the platform compiles a general leaderboard that synthesizes the overall performance of all competing models.