Skip to content

OpenAI Models — Complete Guide

OpenAI offers models across every major AI modality — language, reasoning, code, image, video, audio, and embeddings. The lineup is split into named families, each designed for a different point on the quality-speed-cost curve.


🗂️ Family🎯 Purpose⚡ Speed💰 Cost🌟 Latest
🧠 o-series (reasoning)Deep multi-step reasoningSlowerHighero3-pro
💬 GPT-5.5Frontier language & agentsFastHighGPT-5.5
💬 GPT-4.1 seriesBalanced production languageFastMediumGPT-4.1 / mini / nano
🏠 GPT-oss (open weight)Self-hosted / customVariesInfra costgpt-oss-120b
💻 GPT CodexCoding agentsFastMediumCodex
🖼️ GPT ImageImage generation + editingMediumMediumgpt-image-1.5
🎬 SoraVideo generationSlowHighSora 2
🎙️ Whisper / TranscribeSpeech-to-textFastLowgpt-4o-transcribe
🔊 TTSText-to-speechFastLowgpt-4o-mini-tts
🔍 EmbeddingsText → vector for search/RAGVery fastVery lowtext-embedding-3-large

The o-series models are built for tasks where thinking through a problem step-by-step matters more than raw speed. They use extended internal “thinking” before producing an answer.

🔬 o3-pro — Maximum Reasoning
🔧 Property📋 Detail
🏷️ ClassReasoning — highest compute
🎯 PurposeHard problems where consistency matters most
📄 Best forScientific analysis, complex architecture, production debugging

✅ Use when:

  • 🧩 The task is genuinely hard and ambiguous
  • 💸 Failure would be expensive and hard to undo
  • 🔬 You need maximum reasoning consistency, not just a fast answer

💡 Example:

“Analyse the security implications of this authentication flow, identify the attack vectors, and propose a fix that doesn’t break existing sessions.”

🧠 o3 — Flagship Reasoning
🔧 Property📋 Detail
🏷️ ClassReasoning — flagship
🎯 PurposeMath, science, coding, and multi-step analysis
📄 ContextLarge context window

✅ Use for:

  • 🔢 Advanced mathematics and STEM problems
  • 💻 Complex coding tasks with multiple constraints
  • 🏗️ Architecture and system design decisions
  • 📊 Data analysis requiring multi-step reasoning

💡 Example:

“Design a rate-limiting strategy for an API that handles 50k RPM with burst tolerance, tenant isolation, and graceful degradation under load.”

⚡ o4-mini — Efficient Reasoning
🔧 Property📋 Detail
🏷️ ClassReasoning — compact and efficient
🎯 PurposeFast reasoning on math, code, and visual tasks
🏆 BenchmarksBest-performing on AIME 2024 and AIME 2025

✅ Use for:

  • 🔢 Math and logic tasks at lower cost than o3
  • 👁️ Visual reasoning tasks
  • 💻 Coding problems that need reasoning but not maximum depth
  • ⚡ High-volume reasoning tasks where o3 would be too expensive

GPT-5.5 is OpenAI’s top general-purpose language model. It sits above the GPT-4.1 family for depth of language understanding, instruction-following, and agent workflows.

🔧 Property📋 Detail
🏷️ Model variantsGPT-5.5 · GPT-5.5 mini · GPT-5.5 nano
🎯 PurposeFrontier general reasoning, coding, planning
🌐 AccessOpenAI API · ChatGPT
🤖 Variant🎯 Best For
💬 GPT-5.5Hard analysis, complex coding, agent workflows requiring depth
⚖️ GPT-5.5 miniBalanced — lower latency and cost than the full model
⚡ GPT-5.5 nanoFast, lightweight — classification, routing, extraction

📝 GPT-4.1 Series — Production Workhorses

Section titled “📝 GPT-4.1 Series — Production Workhorses”

The GPT-4.1 family introduced a major expansion: up to 1 million token context alongside improved instruction-following and coding.

💬 GPT-4.1 — Long-Context Flagship
🔧 Property📋 Detail
📄 Context windowUp to 1,000,000 tokens
🎯 FocusInstruction-following · coding · long document analysis
🌐 AccessOpenAI API · ChatGPT

✅ Best for:

  • 📄 Entire codebases or large documents in a single context
  • 💻 Complex coding with many files as input
  • 🔄 Long multi-turn conversations
  • 🏗️ Detailed technical specification analysis
⚖️ GPT-4.1-mini — Balanced
🔧 Property📋 Detail
📄 Context windowUp to 1,000,000 tokens
🎯 FocusSpeed + quality balance
💰 CostLower than GPT-4.1

✅ Best for:

  • 💬 Production chat assistants
  • 📝 Summarisation and drafting
  • 🔄 High-volume tasks needing good quality without the top price
⚡ GPT-4.1-nano — Fastest and Cheapest
🔧 Property📋 Detail
📄 Context windowUp to 1,000,000 tokens
🎯 FocusMaximum speed and minimum cost
💰 CostLowest in the GPT-4.1 family

✅ Best for:

  • 🏷️ Classification, labelling, routing
  • 🔎 Simple extraction from text
  • ⚡ High-volume background processing
  • 📱 Edge-like latency requirements via API

OpenAI’s first open-weight model, released under the Apache 2.0 licence.

🔧 Property📋 Detail
⚙️ Parameters117 billion (MoE — 5.1B active at inference)
📜 LicenceApache 2.0 — free to use and modify
🏃 DeploymentSelf-hosted · any third-party inference provider

✅ Best for:

  • 🏠 On-premise deployment where data cannot leave the organisation
  • 🔧 Fine-tuning on proprietary data
  • 💰 Cost savings at very large inference scale
  • 🏢 Organisations with compliance requirements preventing cloud API use

🖼️ GPT Image — Image Generation and Editing

Section titled “🖼️ GPT Image — Image Generation and Editing”
🖼️ GPT Image 1 / gpt-image-1.5
🔧 Property📋 Detail
🏷️ ModelGPT Image 1 (gpt-image-1.5 is the latest snapshot)
🎯 PurposeText-to-image generation + image editing
📜 NoteDALL-E is the older brand — use GPT Image for modern workflows

✅ Use for:

  • 🖼️ Blog thumbnails and marketing visuals
  • 🎨 Product images and concept art
  • ✏️ Image editing and inpainting
  • 📱 UI mockups and visual prototyping

⚠️ Note: DALL-E model snapshots were deprecated from the API in May 2026.


🎬 Sora / Sora 2 — Text-to-Video
🔧 Property📋 Detail
🏷️ ModelsSora · Sora 2
🎯 PurposeHigh-quality video generation from text prompts
⚠️ API StatusSora 2 Video API deprecated September 2026

✅ Use for:

  • 🎬 Short video clips for marketing or content
  • 🎨 Creative concept visualisation
  • 📱 Social media video content

⚠️ Note: Check current API availability — Sora 2 Video API is being deprecated from the API. The product (ChatGPT) may retain access.


🎙️ Audio Models — Speech-to-Text and Text-to-Speech

Section titled “🎙️ Audio Models — Speech-to-Text and Text-to-Speech”
🎙️ Whisper and gpt-4o-transcribe
🤖 Model↔️ Direction🎯 Best For
whisper-1🎙️ Audio → 📄 textStandard transcription, established pipelines
gpt-4o-transcribe🎙️ Audio → 📄 textHigher accuracy, modern transcription
gpt-4o-mini-transcribe🎙️ Audio → 📄 textFast, cost-efficient transcription

✅ Use for:

  • 📋 Meeting and call transcription
  • 🎬 Subtitle generation
  • 🔍 Audio content indexing for search
  • 📞 Call centre analytics
🔊 Text-to-Speech — gpt-4o-mini-tts
🔧 Property📋 Detail
🏷️ Modelgpt-4o-mini-tts
🎯 PurposeGenerate spoken audio from text
🌐 AccessOpenAI API

✅ Use for:

  • 🤖 Voice agent responses
  • 📖 Narration and accessibility
  • 📞 IVR systems and customer support bots

🔍 Embedding Models — Semantic Search and RAG

Section titled “🔍 Embedding Models — Semantic Search and RAG”
🔍 text-embedding-3-large and text-embedding-3-small
🤖 Model📐 Dimensions🎯 Best For
text-embedding-3-largeUp to 3,072High-precision RAG and semantic search
text-embedding-3-smallUp to 1,536Cost-efficient high-volume search

✅ Use for:

  • 🔍 Semantic search and document retrieval
  • 🤖 RAG pipelines — embed documents, retrieve context, pass to LLM
  • 🔄 Duplicate and near-duplicate detection
  • 💡 Content recommendation engines

⚠️ Embed documents at index time. Embed queries at search time. Never ask the embedding model to answer a question.


🎯 Task✅ Best Model💡 Why
🔬 Hard scientific or logic problemo3 or o3-proExtended reasoning for difficult analysis
💻 Complex multi-file coding tasko3 or GPT-4.1Reasoning + large context for whole-repo tasks
🚨 Production incident debuggingo3-proMaximum reasoning consistency
📄 Long document or large codebase (1M tokens)GPT-4.1Up to 1M token context window
💬 Everyday chat or writing assistantGPT-4.1-mini or GPT-5.5-miniGood quality at reasonable cost
🏷️ Classification, routing, extractionGPT-4.1-nano or GPT-5.5-nanoCheapest and fastest for simple tasks
🖼️ Generate an imagegpt-image-1.5Current OpenAI image generation model
🎙️ Transcribe audiogpt-4o-transcribeBest accuracy for speech-to-text
🔍 Semantic search or RAG retrievaltext-embedding-3-largeHigh-quality vectors for precision retrieval
🏠 Self-hosted / on-premisegpt-oss-120bOpen-weight, Apache 2.0, runs anywhere

🏷️ Platform🤖 Models Available
🟢 OpenAI API (platform.openai.com)All models
💬 ChatGPTGPT-5.5, GPT-4.1, o3, o4-mini, GPT Image, Sora
🔷 Azure OpenAI ServiceGPT-4.1, o3, Whisper, embeddings
🏠 Self-hostedgpt-oss-120b (Apache 2.0)
🔀 Third-party inference providersgpt-oss-120b