Anthropic Leads the AI Ranking for July 2026, OpenAI Falls Behind

Anthropic at the Top of the Overall Ranking

For several months, Anthropic has established itself as a major player in the field of artificial intelligence, gradually gaining market share from OpenAI. In July 2026, this trend continues with Anthropic maintaining its dominance in the overall ranking, although there are a few surprises in specific segments such as web development.

In June, Anthropic had already managed to secure the top five positions in the overall ranking. This situation persists in July, with Claude Fable 5 holding the first position, closely followed by the "thinking" versions of Claude Opus 4.6 and 4.7. The standard versions of these models complete the top 5. The journey of Claude Fable 5 has been eventful since its launch on June 9, being the first public model in the Mythos class. It was briefly suspended by the U.S. Department of Commerce on June 12 for all foreign users, before being reinstated on July 1 after Washington lifted the restrictions.

In the rest of the ranking, Meta continues its quiet progress. Its Muse Spark models are positioned in sixth and seventh place, surpassing Google, Moonshot, and OpenAI, which find themselves in eighth, ninth, and tenth positions respectively.

Here are the 10 best-performing language models in July 2026, across all tasks:

• Claude Fable 5
• Claude Opus 4.6 "thinking"
• Claude Opus 4.7 "thinking"
• Claude Opus 4.6
• Claude Opus 4.7
• Muse Spark 1.1
• Muse Spark
• Gemini 3 Pro
• kimi-k3
• GPT-5.6 Sol (xHigh)

New Rankings by Category

This month, Arena introduced a new category dedicated to agentic AI, adding an additional dimension to the traditional thematic rankings. Here are the main trends observed in July 2026:

• Agentic AI: Anthropic and OpenAI are competing for the top spots in this new category, with Anthropic slightly ahead. Claude Fable 5 (High) leads over GPT-5.6 Sol (xHigh) and Claude Opus 4.8 "thinking".

• Web Development: After a long period of dominance by Anthropic, Moonshot surprises with kimi-k3 taking the first place, ahead of Claude Fable 5 and GPT-5.6 Sol.

• Image Analysis: Anthropic regains control with Claude Fable 5 at the top, followed by several variations of Claude Opus. Meta's Muse Spark ranks sixth, ahead of Gemini 3 Pro.

• Document Analysis: Anthropic continues to dominate this category, with Claude Opus 4.6 "thinking" in the lead, closely followed by Claude Fable 5 and Claude Opus 4.6. OpenAI manages to sneak into the top 10.

• Image Generation: OpenAI maintains a notable lead with GPT-Image-2 (medium), followed by Reve 2.1 and Meta's Muse Image. Google places four variants of Nano Banana in the top 10.

• Web Search: Claude Opus 4.6 "search" remains at the top of a highly contested category, with GPT-5.5 "search" in second place. Baidu's Ernie-5.1 and two Gemini variants from Google show that several players remain competitive behind Anthropic.

Arena's Ranking Methodology

Arena relies on a continuous evaluation method based on user feedback. The platform organizes anonymized duels between different models, each subjected to the same prompt. Users are then invited to choose the model that best responds to the request. Each model receives an Elo score, similar to that used in chess, which evolves based on results: a model gains points by defeating a higher-ranked competitor and loses points against a lower-ranked opponent.