What LLMs Can Do
Exploring the capabilities of Large Language Models across different domains
Chat
mature
90
/100
Advanced conversational abilities with strong reasoning and multilingual support (Based on Google Sheets MMLU scores)
Subsections
General Conversation
mature
90
/100
Natural dialogue with high context awareness and multilingual capabilities
| Feature |  Claude 3 Opus |  GPT-4 Turbo |  Gemini 1.5 Pro |
|---|---|---|---|
Context Awareness Maintains context across conversation turns | Full 200k token context | Full 128k token context | Full 1M token context |
Multilingual Support for multiple languages | Full 100+ languages | Full All major languages | Full All major languages |
Reasoning And Problem Solving
mature
85
/100
Complex reasoning, mathematical problem-solving, and logical deduction
| Feature | Claude 3 Opus | GPT-4 Turbo | Gemini 1.5 Pro |
|---|---|---|---|
Mathematical Reasoning Solving complex mathematical problems | Full Highest MMLU math | Full 90th percentile UBE | Partial MMLU-Pro 76.4% |
Logical Deduction Step-by-step logical reasoning | Full 93.7% HumanEval score | Full Advanced reasoning | Full Strong reasoning |
Top Players
Claude 3 Opus
Anthropic
- 93.7% HumanEval score
- 200k token context
- Best-in-class reasoning
- Highest MMLU scores
GPT-4 Turbo
OpenAI
- 128k context window
- 90th percentile UBE
- Advanced reasoning
- Comprehensive API
Gemini 1.5 Pro
Google
- 1M token context
- Strong performance
- Video understanding
- Efficient processing
Coding
mature
80
/100
Advanced code generation and analysis with specialized tools and models (Based on SWE-bench data)
Subsections
Code Completion
mature
80
/100
Real-time code suggestions with multi-language support and context awareness
| Feature | GitHub Copilot |  Cursor |  Windsurf |  All Hands |
|---|---|---|---|---|
Real Time Suggestions Intelligent code suggestions in real-time | Full Real-time completions | Full Context-aware suggestions | Full Smart completions | Partial Basic suggestions |
Multi Language Support for multiple programming languages | Full 40+ languages | Full All major languages | Partial Major languages | Partial Major languages |
Context Aware Understanding project context and dependencies | Full Full repository context | Full Codebase understanding | Full Full repo context | Full Team context aware |
Code Modification
emerging
75
/100
Code refactoring and optimization with AI-powered improvements
| Feature | GitHub Copilot | Cursor | Windsurf | All Hands |
|---|---|---|---|---|
Refactoring Automated code refactoring and improvements | Partial Basic refactoring | Full AI-powered edits | Full Smart transformations | Partial Collaborative improvements |
Bug Fixing Automated bug detection and fixing | Full Advanced bug detection | Full Smart bug fixing | Full Automated fixes | None |
Pr Review
emerging
70
/100
Automated code review and analysis of pull requests
| Feature | GitHub Copilot | Cursor | Windsurf | All Hands |
|---|---|---|---|---|
Automated Review AI-powered code review comments | Partial Through Copilot Chat | Partial Basic review | Full Comprehensive review | Full Team-focused review |
Security Analysis Identifying security issues and vulnerabilities | Full Security features | None | Partial Basic security checks | Partial Basic security review |
Agentic Programming
early
65
/100
Autonomous development and task planning capabilities
| Feature | GitHub Copilot | Cursor | All Hands |
|---|---|---|---|
Autonomous Development Independent code development capabilities | Partial Basic automation | Full Advanced autonomous coding | Full Full autonomous development |
Task Planning Planning and organizing coding tasks | Partial Basic planning | Full Advanced task planning | Full Team-aware planning |
Top Players
GitHub Copilot
GitHub/OpenAI
- Real-time code suggestions
- IDE integration
- Context-aware completions
Cursor
Cursor
- Context-aware completions
- AI-powered edits and refactoring
- Basic code review capabilities
Windsurf
Windsurf AI
- Context-aware suggestions
- Smart code transformations
- Automated code review
All Hands
All Hands
- Team-aware code suggestions
- Collaborative code improvements
- Team-focused code review
Multimodal Understanding
emerging
75
/100
Processing and understanding multiple types of input (text, images, audio)
Subsections
Vision Analysis
mature
80
/100
Understanding and analyzing visual content
| Feature | Gemini Ultra | Gemini 1.5 Pro | GPT-4V | Claude 3 Opus |
|---|---|---|---|---|
Image Understanding Detailed analysis of image content and context | Full Advanced multimodal | Full Advanced visual processing | Full High-res image analysis | Full Detailed visual comprehension |
Ocr Capabilities Text extraction from images | None | Partial Basic text extraction | Full Advanced OCR | Full Mathematical notation |
Audio Processing
emerging
70
/100
Processing and understanding audio input
| Feature | Gemini Ultra | Gemini 1.5 Pro |
|---|---|---|
Speech Recognition Converting speech to text with high accuracy | Full Advanced speech recognition | Full Multiple languages |
Cross Modal Reasoning
emerging
75
/100
Understanding relationships between different types of input
| Feature | Gemini 1.5 Pro | GPT-4V | Claude 3 Opus |
|---|---|---|---|
Visual Reasoning Complex reasoning about visual information | Full Cross-modal understanding | Full Visual reasoning | Full Diagram analysis |
Multimodal Generation Generating responses combining multiple modalities | Full Text and image generation | Partial Text-based responses only | Partial Text-based responses only |
Top Players
Gemini Ultra
Google
- 91.8 MMLU score
- Advanced multimodal
- Cross-modal reasoning
Gemini 1.5 Pro
Google
- 1M token context
- Video understanding
- Advanced visual processing
GPT-4V
OpenAI
- High-resolution image analysis
- Visual reasoning with GPT-4 Turbo
- Advanced OCR capabilities
Claude 3 Opus
Anthropic
- 200k token context
- 93.7% HumanEval score
- Advanced visual analysis
- Mathematical diagram comprehension