The Great AI EQ Battle: 2025's Latest EQ-Bench Rankings Revealed, Who is the Most Emotionally Intelligent Language Model?

AI is no longer just a cold machine. The latest EQ-Bench 3 emotional intelligence evaluation rankings are out, and the results might surprise you. This article will delve into this list, examining the true performance of top models like Horizon-Alpha, Kimi, GPT-5, and Gemini in ‘reading the room,’ and explore why emotional intelligence is becoming the next key battleground in AI development.

Have you ever wondered, when we chat with an AI, what do we expect besides accurate answers? Perhaps a feeling of being understood, a warm response, or even a tacit understanding that can ‘read the room.’ Frankly, this is ‘Emotional Intelligence’ (EQ), and it’s quietly becoming a new dimension for judging the quality of an AI model.

Recently, the authoritative AI emotional intelligence evaluation platform EQ-Bench released its latest third edition leaderboard. This list is like the ‘EQ final exam’ of the AI world, examining the ability of major models to handle complex emotional interactions through challenging role-playing scenarios.

So, in 2025, which model truly understands the ‘human heart’? The results might not be what you think.

What is EQ-Bench? And Why is it So Important?

Before we unveil the list, we need to talk about what EQ-Bench is. Simply put, it’s not a platform for testing an AI’s calculation or programming abilities, but is specifically designed to measure the performance of Large Language Models (LLMs) in emotional communication.

The evaluation method is very special: it has the model participate in some tricky, emotionally charged simulated conversations, and then another high-performance model (currently Sonnet 3.7 serves as the judge) scores it from multiple dimensions such as empathy, insight, and social acuity. Finally, through a Elo rating system similar to chess competitions, a comprehensive emotional intelligence score is given.

Why is this important? Because as AI integrates into our daily lives, whether as a work assistant, learning partner, or life companion, its emotional intelligence will directly determine whether our experience is smooth and pleasant, or full of frustration. A high-EQ AI can truly become our capable assistant, not just a talking calculator.

Latest AI Emotional Intelligence Rankings for August 2025 (Elo Score)

Alright, here comes the main event. Let’s take a look at this latest list as of August 14, 2025. Please note that the higher the Elo score, the stronger the comprehensive emotional intelligence performance. As for the colorful ability scores on the side, they are not included in the total score, but they give us a glimpse into the unique ‘personality’ of each model.

Source: EQ-Bench Official Website

Highlights and Reflections from the Leaderboard: Who is the Unexpected Dark Horse?

After seeing this list, are you also a bit surprised? Here are a few findings worthy of our deep thought:

1. A New King is Crowned: Who is Horizon-Alpha? The top spot is no longer held by the giants we are familiar with. A model named horizon-alpha has taken the crown with a slight advantage, boasting an Elo score of 1568. The emergence of this dark horse proves how fierce the competition in the AI field is, with new challengers always ready to disrupt the landscape.

2. Kimi is Hot on its Heels Kimi-K2-Instruct from China is in second place with a high score of 1565, only 3 points behind the leader. Looking at the ability heat map, Kimi scored an astonishing 9.6 in Insight, Empathy, and Analytic, showing its outstanding performance in deeply understanding and responding to user emotions.

3. Has GPT-5’s Emotional Intelligence ‘Regressed’? This might be the most surprising point. The latest gpt-5-chat-latest-2025-08-07 has an Elo score of 1357, which is actually lower than the chatgpt-4o-latest-2025-03-27 (1370 points) released a few months ago. This raises an interesting question: does the iteration and update of a model necessarily bring an improvement in emotional intelligence? Perhaps the new model is stronger in logical reasoning or coding ability, but in terms of emotional delicacy, it is not as pleasing as the old version. This reminds us that the ‘progress’ of AI is multi-dimensional and cannot be judged by a single indicator.

4. Not Just a Score, but a Showcase of ‘Personality’ If you look closely at the heat map, you will find that each model has its own ‘personality.’ For example, some models may have a high Warm score, like a friendly friend; others have outstanding Analytic abilities, like a calm strategist. And some models have a higher score in Moralising, which means it may prefer to ’educate’ users, which can be a bit annoying in some situations. This is the charm of EQ-Bench, it allows us to see the diverse personality profiles of AI.

Interpreting EQ-Bench: What Qualities Do High-EQ AIs Possess?

The scoring of EQ-Bench is not just a number; it has a complete evaluation system behind it, mainly revolving around eight core dimensions, while also observing some non-scoring traits.

Core Scoring Dimensions:

• Demonstrated empathy: The ability to recognize, understand, and share the feelings of others.
• Pragmatic EI: The ability to apply emotional intelligence to solve practical problems.
• Depth of insight: The ability to provide profound, novel perspectives and identify potential problems.
• Social dexterity: The ability to handle social interactions with ease.
• Emotional reasoning: The ability to conduct logic-based thinking based on emotions.
• Appropriate validation and/or challenge: Knowing when to give affirmation and when to offer a different perspective.
• Message tailoring: Adjusting the communication style according to the audience and context.
• Overall EQ: The overall emotional intelligence performance.

‘Personality’ Traits for Reference Only:

• Humanlike: The naturalness and human-like degree of the response.
• Assertive: The ability to confidently set boundaries when needed.
• Warm: A friendly, approachable, and easy-to-talk-to tone.
• Compliant: Following instructions or agreeing to the user’s wishes.

Conclusion: The Future of AI Begins with the ‘Heart’

This EQ-Bench leaderboard reveals an important trend in AI development: the technological race is shifting from a simple ‘IQ’ competition to a more complex ‘EQ’ contest.

A high-EQ AI can not only complete tasks more efficiently, but also build emotional connections and trust with humans. In the future, when we choose AI services, we may be like choosing friends, not only looking at how smart it is, but also valuing whether it ‘understands me.’

This great AI EQ battle has just begun. What surprises will the next leaderboard bring? Let’s wait and see.