AI Models Guide
Choose the right model for every task
Lookinglass gives you access to 27 OpenAI models (12 GPT chat models, 7 GPT-5 reasoning models, and 8 O-series models) organized by model family and use case. The app includes powerful filtering to help you find the perfect model—filter by reasoning capability, web search support, image input, input caching, Smart Savings availability, context window size, and more.
- Default: gpt-4.1-mini (gpt-4.1 series) - Start here for everyday tasks. Solid performance for writing, coding, and analysis with 1M token context
- Latest Generation: gpt-5 series (7 reasoning models) - Newest model family (December 2025). Reasoning models with customizable AI power and response detail. Note: gpt-5-chat-latest, gpt-5.1-chat-latest, and gpt-5.2-chat-latest are chat models (see GPT chat models section)
- Long Conversations: gpt-4.1 (3 models) - Best for long threads where significant context is needed (up to 1M tokens). gpt-4.1 and gpt-4.1-mini support web search
- Proven & Reliable: gpt-4o series (3 models) - OpenAI's former flagship model. Battle-tested, widely trusted, and still excellent for daily work. Supports web search
- Hard Problems: o3 series (3 models) - Best reasoning models. Use when you need step-by-step problem solving for math, logic, or complex coding challenges. o3 supports web search
- Deep Research: o3-deep-research & o4-mini-deep-research (2 models) - Specialized for intensive research and analysis. Perfect for comprehensive investigations requiring up-to-date information
- Alternative Reasoning: o1 series (2 models) - Previous generation reasoning. OpenAI recommends o3, but o1 remains available for users who prefer its approach
- Budget Reasoning: o4-mini (1 model) - Most affordable reasoning model with web search support, good for basic reasoning tasks
- Legacy: gpt-4 & gpt-3.5 (3 models) - Legacy models, rarely the best choice
This guide helps you choose the right model and settings for your needs.
Understanding the Basics
Tokens & Pricing
Tokens are how AI models measure text. Roughly 1 token ≈ 4 characters, so "Hello world!" is about 3 tokens. Both your messages and the AI's responses are measured in tokens.
- 100 tokens ≈ 75 words (one paragraph)
- 1,000 tokens ≈ 750 words (one page)
Every model has a rate per million tokens. Your cost is simple: tokens used × model rate. Lookinglass shows you the maximum possible cost before you send each message—no surprises, ever.
AI Workspace
Your AI's workspace is its total "working memory"—the maximum amount of information it can process in a single request. Think of it as the AI's desk space that must accommodate everything it needs to work with.
- Your current message what you're asking
- Context window recent conversation history
- Permanent memory your preferences and style
- Its response the answer it generates
- Internal reasoning for reasoning models
Bigger context = more conversation history the AI remembers.
Find Your Perfect Model
Lookinglass includes powerful filtering to help you find the right model for any task. Browse by category or filter by specific capabilities.
Model Categories
Models are organized into four categories by use case, making it easy to find what you need:
- Everyday Models: Balanced performance for daily tasks—your go-to models for most work
- Lightweight Models: Mini and nano variants optimized for speed and cost—perfect when you need quick answers
- Advanced Models: Pro models and deep research variants for the most demanding tasks
- Legacy Models: Older generation models maintained for compatibility
Filter by Capabilities
Narrow down models by specific features you need:
- Reasoning: Models that think step-by-step through complex problems
- Web Search: Models that can search the internet for up-to-date information (18 models)
- Image Input: Models that accept and analyze images (up to 5 per message)
- Input Caching: Models that support cached tokens for faster, cheaper follow-up messages
- Smart Savings: Models with 50% discount option for non-urgent tasks
Filter by Settings
Find models that support the controls you want:
- Creativity Level: Adjust temperature for more creative or focused responses
- Response Detail Level: Control how concise or thorough responses are
- AI Power Level: Set reasoning effort from off (turns model into non-reasoning with lower latency) to light, standard, power, ultimate, or max (gpt-5.2)
Filter by Context Size
Choose models by minimum context window:
- Any: All 27 models available
- 128K+: Models with at least 128,000 tokens (25 models)
- 200K+: Models with at least 200,000 tokens (18 models)
- 400K+: Models with at least 400,000 tokens (10 models)
- 1M+: Models with 1 million token context (3 models)
GPT Chat Models (12 Available)
General-purpose conversation and task completion models, perfect for everyday AI interactions. Most models support image input (up to 5 images per message), except gpt-3.5-turbo and gpt-4.
gpt-5 Chat Models (3 Models)
Non-reasoning chat variants of the gpt-5 family. gpt-5-chat-latest, gpt-5.1-chat-latest, and gpt-5.2-chat-latest provide the latest knowledge (August 31, 2025 cutoff for gpt-5.2-chat-latest; September 30, 2024 for others) without reasoning capabilities. Perfect for everyday conversations and tasks that don't require step-by-step thinking. All support image input (up to 5 images per message) and web search. Learn more from OpenAI.
| Model: | gpt-5-chat-latest | gpt-5.1-chat-latest | gpt-5.2-chat-latest |
| Intelligence: | ●●●○○ | ●●●○○ | ●●●○○ |
| Speed: | ⚡⚡⚡○○ | ⚡⚡⚡○○ | ⚡⚡⚡○○ |
- models: gpt-5-chat-latest, gpt-5.1-chat-latest, gpt-5.2-chat-latest
- best for: Latest knowledge, everyday conversations, tasks that don't require reasoning; gpt-5.2-chat-latest best for coding and agentic tasks
- context: 128K tokens
- image support: yes (up to 5 images per message)
- web search: yes
- smart savings: no
- starting from: $0.0002 per message
gpt-4o Series (3 Models)
Released May 13, 2024, gpt-4o ("omni") was OpenAI's first truly multimodal model, processing text, audio, and vision natively. Featured dramatically faster responses, superior multilingual capabilities (50+ languages), and free access that democratized advanced AI. Anecdotal user feedback suggests its communication style remains preferred by many users over newer models. gpt-4o and gpt-4o-mini support web search. Technical deep dive (system card).
| Model: | chatgpt-4o-latest | gpt-4o | gpt-4o-mini |
| Intelligence: | ●●●○○ | ●●●○○ | ●●○○○ |
| Speed: | ⚡⚡⚡○○ | ⚡⚡⚡○○ | ⚡⚡⚡⚡○ |
- models: chatgpt-4o-latest, gpt-4o, gpt-4o-mini
- best for: proven reliable choice for daily work, battle-tested for all tasks
- context: 128K tokens
- image support: yes (up to 5 images per message)
- web search: yes (gpt-4o, gpt-4o-mini only)
- smart savings: no
- starting from: $0.00002 per message
gpt-4.1 Series (3 Models)
Released April 14, 2025, gpt-4.1 introduced the industry's largest context window at 1M tokens—enough to analyze entire codebases or multiple books simultaneously. Designed for tasks requiring significant context, long-form content analysis, and complex multi-document reasoning. Also features improved tool-calling and more precise instruction following compared to previous generations. gpt-4.1 and gpt-4.1-mini support web search.
| Model: | gpt-4.1 | gpt-4.1-mini | gpt-4.1-nano |
| Intelligence: | ●●●●○ | ●●●○○ | ●●○○○ |
| Speed: | ⚡⚡⚡○○ | ⚡⚡⚡⚡○ | ⚡⚡⚡⚡⚡ |
- models: gpt-4.1, gpt-4.1-mini, gpt-4.1-nano
- best for: tasks requiring significant context, long conversations, and analyzing large amounts of text
- context: 1M tokens
- image support: yes (up to 5 images per message)
- web search: yes (gpt-4.1, gpt-4.1-mini only)
- smart savings: no
- starting from: $0.00001 per message
gpt-4 Series (2 Models)
Released March 14, 2023, gpt-4 was OpenAI's breakthrough model that dramatically improved capabilities over gpt-3.5, introducing multimodal processing, larger context windows, and significantly better performance on challenging tasks. Now superseded by newer models. Rarely the best choice unless you need compatibility with existing workflows built on the original gpt-4 architecture.
| Model: | gpt-4 | gpt-4-turbo |
| Intelligence: | ●●○○○ | ●●○○○ |
| Speed: | ⚡⚡⚡○○ | ⚡⚡⚡○○ |
- models: gpt-4 (8K), gpt-4-turbo (128K)
- best for: Legacy model, supported for compatibility with existing workflows and testing
- context: 8K - 128K tokens
- image support: no (gpt-4), yes (gpt-4-turbo, up to 5 images per message)
- smart savings: no
- starting from: $0.001 per message
gpt-3.5 Series (1 Model)
Released March 15, 2022, gpt-3.5 powered the initial ChatGPT launch in November 2022, sparking the modern AI boom. Preceded by gpt-3 and succeeded by gpt-4 (March 2023). While historically groundbreaking, newer models have far surpassed it. Mainly useful for testing workflows. Rarely the best choice for any serious task.
| Model: | gpt-3.5-turbo |
| Intelligence: | ●○○○○ |
| Speed: | ⚡⚡○○○ |
- models: gpt-3.5-turbo
- best for: Legacy model, supported for compatibility with existing workflows and testing
- context: 16K tokens
- image support: no
- smart savings: no
- starting from: $0.00005 per message
GPT-5 Reasoning Models (7 Available)
Advanced reasoning models from the gpt-5 family that "think" step-by-step through complex problems. All support image input (up to 5 images per message), web search, and customizable AI power levels.
gpt-5 Series (5 Models)
Released August 7, 2025, gpt-5 represents OpenAI's latest generation with knowledge cutoff September 30, 2024. Includes five reasoning-capable models (gpt-5, gpt-5.1, gpt-5-pro, gpt-5-mini, gpt-5-nano) with AI Power Level controls—though gpt-5-pro always operates at maximum power. Unique features: customizable AI power level (thinking time), response detail level, and web search support. gpt-5.1 introduces a special "none" AI power level that disables reasoning, effectively turning it into a non-reasoning model. Multiple size options let you balance speed, quality, and cost. Learn more from OpenAI.
| Model: | gpt-5-pro | gpt-5 | gpt-5.1 | gpt-5-mini | gpt-5-nano |
| Intelligence: | ●●●●● | ●●●●○ | ●●●●○ | ●●●○○ | ●●○○○ |
| Speed: | ⚡○○○○ | ⚡⚡⚡○○ | ⚡⚡⚡⚡○ | ⚡⚡⚡⚡○ | ⚡⚡⚡⚡⚡ |
- models: gpt-5, gpt-5.1, gpt-5-pro, gpt-5-mini, gpt-5-nano
- best for: Latest knowledge with reasoning capabilities. Use for complex tasks requiring step-by-step thinking
- context: 400K tokens
- image support: yes (up to 5 images per message)
- web search: yes
- smart savings: yes (gpt-5, gpt-5-mini, and gpt-5-nano)
- special feature: gpt-5.1 supports "none" AI power level that disables reasoning
- starting from: $0.00001 per message
gpt-5.2 Series (2 Models)
The best model for coding and agentic tasks across industries. Released with knowledge cutoff August 31, 2025, gpt-5.2 and gpt-5.2-pro are optimized specifically for coding and agentic workflows. Features expanded AI power levels including "max" for ultimate thinking depth. gpt-5.2-pro always operates at maximum power. Both support image input (up to 5 images per message), web search, and customizable response detail. Learn more from OpenAI.
| Model: | gpt-5.2 | gpt-5.2-pro |
| Intelligence: | ●●●●○ | ●●●●○ |
| Speed: | ⚡⚡⚡⚡○ | ⚡○○○○ |
- models: gpt-5.2, gpt-5.2-pro
- best for: Coding and agentic tasks. Use for software development, code review, and autonomous agent workflows
- context: 400K tokens
- image support: yes (up to 5 images per message)
- web search: yes
- smart savings: yes (both models)
- special feature: gpt-5.2 supports "max" AI power level for ultimate thinking depth
- starting from: $0.0002 per message
O-Series Reasoning Models (8 Available)
Advanced reasoning models that "think" step-by-step through complex problems, perfect for mathematics, logic, and detailed analysis. Includes specialized deep research variants optimized for comprehensive investigations. Most support image input (up to 5 images per message), except o3-mini.
o3 Series (3 Models)
Released April 16, 2025, o3 is the successor to o1 with improved reasoning capabilities and better performance on complex multi-step problems. Features configurable thinking depth and significantly outperforms o1 on reasoning tasks. Notable for its ability to process both text and images during its chain-of-thought phase, including analyzing whiteboard sketches. Perfect for advanced math, programming challenges, and scientific problems. o3 supports web search. Learn more from OpenAI.
| Model: | o3-pro | o3 | o3-mini |
| Intelligence: | ●●●●● | ●●●●● | ●●●●○ |
| Speed: | ⚡○○○○ | ⚡○○○○ | ⚡⚡⚡○○ |
- models: o3, o3-pro, o3-mini
- best for: complex reasoning, research-level problems, mathematical proofs
- AI power levels: standard, power, ultimate
- context: 200K tokens
- image support: yes (o3, o3-pro). Note: o3-mini does not support images
- web search: yes (o3, o3-pro only)
- smart savings: yes (o3)
- starting from: $0.0002 per message
o1 Series (2 Models)
Released September 12, 2024, o1 pioneered OpenAI's "reasoning" paradigm, spending additional time thinking before answering to solve complex problems. The full o1 model was released December 5, 2024. Previously code-named "Strawberry" and "Q*". Performs at PhD level on physics, chemistry, and biology benchmarks. Succeeded by o3. Learn more from OpenAI.
| Model: | o1-pro | o1 |
| Intelligence: | ●●●●○ | ●●●●○ |
| Speed: | ⚡○○○○ | ⚡○○○○ |
- models: o1, o1-pro
- best for: mathematical problems, logical reasoning, academic work
- AI power levels: standard, power, ultimate
- context: 200K tokens
- image support: yes (up to 5 images per message)
- smart savings: no
- starting from: $0.002 per message
o4-mini Series (1 Model)
Released April 16, 2025 alongside o3, o4-mini is a smaller, more efficient reasoning model. Unlike earlier O-series models, o4-mini can process both text and images, enabling it to analyze whiteboard sketches during its chain-of-thought phase. Designed for everyday logical tasks at a lower cost than full o3, making reasoning capabilities accessible for educational problems and basic logical challenges. Learn more from OpenAI.
| Model: | o4-mini |
| Intelligence: | ●●●●○ |
| Speed: | ⚡⚡⚡○○ |
- model: o4-mini
- best for: basic reasoning, educational problems, logic puzzles
- AI power levels: standard, power, ultimate
- context: 200K tokens
- image support: yes (up to 5 images per message)
- web search: yes
- smart savings: yes
- starting from: $0.0001 per message
Deep Research Models (2 Models)
Specialized variants designed for intensive research and analysis tasks. These models combine powerful reasoning capabilities with web search to gather and synthesize information from across the internet. Perfect for comprehensive investigations, up-to-date research, and tasks requiring evidence from multiple sources. Deep research models take 5-15+ minutes to complete their analysis, making them ideal for non-urgent, thorough research tasks. Learn more from OpenAI.
| Model: | o3-deep-research | o4-mini-deep-research |
| Intelligence: | ●●●●● | ●●●●○ |
| Speed: | ⚡○○○○ | ⚡○○○○ |
- models: o3-deep-research, o4-mini-deep-research
- best for: comprehensive research requiring multiple sources, up-to-date information gathering, deep investigation into complex topics
- reasoning: yes (AI power level not configurable)
- context: 200K tokens
- image support: yes (up to 5 images per message)
- web search: yes
- response time: 5-15+ minutes
- smart savings: no
- starting from: varies based on research depth
Recommended Models
Compare chat and reasoning models by use case
Fastest & Cheapest
Quick answers, simple tasks, basic questions
- recommended: gpt-4.1-nano
- why: lightning fast, lowest cost
- example: "What's the capital of France?" or "Fix this syntax error"
Best Overall (Default)
Most chats, summaries, balanced speed and cost
- recommended: gpt-4.1-mini
- why: 1M token window, fast, affordable
- example: "Summarize this article" or "Help me write an email"
Highest Quality
Advanced work, creative projects, complex conversations
- recommended: chatgpt-4o-latest or gpt-5.2-chat-latest
- why: top non-reasoning models, continuously updated by OpenAI
- example: "Write a detailed business plan" or "Refactor this codebase"
Fastest & Cheapest Reasoning
Quick step-by-step thinking on a budget
- recommended: gpt-5-nano
- why: cheapest reasoning model, 50% off with smart savings
- example: "Explain this concept step-by-step"
Balanced Reasoning
Solid reasoning for everyday problems
- recommended: gpt-5-mini or o4-mini
- why: balanced speed, cost, and quality, 50% off with smart savings
- example: "Walk me through solving this problem"
Advanced Reasoning
Complex math, logic, research-level problems
- recommended: o3 or gpt-5.2
- why: best reasoning outside pro models, 50% off with smart savings
- example: "Solve this calculus problem" or "Debug this algorithm"
Advanced Settings
Creativity Level
Controls how creative or focused the AI's responses are.
- low (0.2): focused, consistent, factual responses
- balanced (1.0): balanced creativity and accuracy (default)
- high (1.5): creative, varied, sometimes unexpected
- availability: gpt-5-chat-latest, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4o, gpt-4o-mini, chatgpt-4o-latest, gpt-4, gpt-4-turbo, and gpt-3.5-turbo. Note: gpt-5.1-chat-latest and gpt-5.2-chat-latest do not support creativity level
AI Power Level (Reasoning Models)
How much time the AI spends thinking through problems.
- none: disables reasoning, turns gpt-5.1 and gpt-5.2 into a non-reasoning model (gpt-5.1 and gpt-5.2 only)
- light: minimal thinking, fastest, lowest cost (gpt-5, gpt-5-mini, and gpt-5-nano)
- standard: quick thinking, good for most tasks
- power: deeper thinking, balanced cost and quality
- ultimate: deep thinking, high cost, highest quality for most models
- max: maximum thinking depth, highest cost (gpt-5.2 only)
- availability: gpt-5, gpt-5.1, gpt-5.2, gpt-5-mini, gpt-5-nano, o1-pro, o1, o3-pro, o3, o3-mini, and o4-mini
Response Detail Level
Controls how much detail the AI includes in its responses.
- concise: brief, to-the-point answers
- balanced: standard level of detail (default)
- detailed: thorough, comprehensive explanations
- availability: gpt-5, gpt-5.1, gpt-5.2, gpt-5-mini, and gpt-5-nano
Billing Plan
Choose between speed and savings based on your needs.
fast & reliable: faster responses, standard pricing
smart savings: 50% off, slight delay (select models only)- availability: gpt-5, gpt-5.2, gpt-5.2-pro, gpt-5-mini, gpt-5-nano, o3, o3-mini, and o4-mini