OPENAI MODEL SELECTION GUIDE · 2026
EN

OpenAI Model Selection Guide

78+ models · multimodal · reasoning · context window · price tiers — Sources: OpenAI official docs · Azure AI Foundry docs · Azure model retirement (synced 2026-08-30)

78Models
9Categories
1.05MMax context
Tier
FlagshipBalancedBudget
Speed
FastestFastStandardSlowSlowest
Modality
TextImageAudioVideoCode
Reasoning
BestDeepStandardBasicFast
Price
Premium$100+ / $500+Expensive$10-100 / $50-500Costly$2-10 / $8-50Moderate$0.5-2 / $2-8Affordable$0.1-0.5 / $0.4-2Cheapest<$0.1 / <$0.4Unit: USD / 1M tokens (input / output)

Quick Picks

Direct recommendation by task type

Chat / Writinggpt-chat-latest
Codinggpt-5.3-codex
Image generationgpt-image-2
Video generationsora-2
Live voice transcriptiongpt-live-transcribe
Text to speechtts-1-hd
Realtime voice chatgpt-realtime-2.1
Deep research / complex reasoningo3-pro
Top reasoninggpt-5.4-pro
Cheapestgpt-5.4-nano
Long-document processinggpt-5.6-sol
Self-hosted deploymentgpt-oss-120b

Naming Conventions

Understand these patterns and instantly grasp any model name

SuffixMeaningExample
Version number (4→5→5.6) Higher number = newer and stronger gpt-4 → gpt-5 → gpt-5.6
Codename (-sol/-terra/-luna) Variant codename within the same generation (e.g. GPT-5.6 family) gpt-5.6-sol, gpt-5.6-cyber
-mini Mid-size, balanced performance and cost gpt-5-mini, o4-mini
-nano Smallest, cheapest and fastest gpt-5-nano, gpt-4.1-nano
-pro Enhanced, more compute and accuracy gpt-5.2-pro, o3-pro
-codex Coding-specialized gpt-5.2-codex
-max Optimized for long tasks gpt-5.1-codex-max
-turbo Speed-optimized version gpt-4-turbo
-hd HD / high-quality version tts-1-hd
-oss-XXb Open-source, XX = parameters in billions gpt-oss-120b
o1 / o3 / o4 Reasoning model family (deep thinking) o3, o4-mini
-deep-research Deep-research specialized, extra-long reasoning o3-deep-research
-preview / -latest Preview / latest (auto-updating) computer-use-preview, gpt-chat-latest
-transcribe / -tts Speech-to-text / text-to-speech gpt-transcribe, gpt-4o-mini-tts
-realtime / -live Realtime two-way audio / live transcription gpt-realtime-2.1, gpt-live-transcribe

Frontier Models

OpenAI's most advanced models, recommended for most tasks. GPT-5.6 is the current generation (released 2026-07), with multi-agent orchestration (preview) and computer use; tool calling must go through the Responses API (Chat Completions and function tools can't be combined unless reasoning_effort=none)

Model IDTierPriceModalityReasoningSpeedContextInputOutputNotes
gpt-5.6-solNEW Flagship
Best Fast 1.05M 922K 128K GPT-5.6 family (codename Sol), multi-agent orchestration (preview)
gpt-5.6-terraNEW Flagship
Deep Fast 1.05M 922K 128K GPT-5.6 family (codename Terra), multi-agent orchestration (preview)
gpt-5.6-lunaNEW Flagship
Standard Fast 1.05M 922K 128K GPT-5.6 family (codename Luna), multi-agent orchestration (preview)
gpt-5.6-cyberNEW Flagship
Best Fast 400K 272K 128K GPT-5.6 family (codename Cyber), $12.5/$75
gpt-daybreak-red-latestNEW Flagship
Best Fast 400K 272K 128K Alias of gpt-5.6-cyber for offensive security research (Daybreak program); separate approval required
gpt-daybreak-blue-latestNEW Flagship
Best Fast 1.05M 922K 128K Alias of gpt-5.6-sol for defensive security work (Daybreak program); separate approval required
gpt-chat-latestNEWPreview Flagship
Basic Fast 400K 272K 128K Continuously updated chat model (i.e. GPT-5.5 Instant); fixed reasoning effort
gpt-5.5 Flagship
Best Fast 1.05M 922K 128K ~922K token effective budget via Responses API
gpt-5.5-proNEW Flagship
Best Slowest 1.05M 922K 128K Enhanced reasoning, Responses API only (incl. Batch); no cached-input discount
gpt-5.4 Flagship
Best Standard 1.05M 922K 128K Supports computer use; great for coding and agent tasks
gpt-5.4-proPRO Flagship
Best Slowest 1.05M 922K 128K Enhanced reasoning, Responses API only, unlimited compute
gpt-5.4-mini Balanced
Deep Fast 400K 272K 128K Lightweight, supports computer use
gpt-5.4-nano Budget
Standard Fast 400K 272K 128K Fastest, cheapest GPT-5.4
gpt-5.3-chatDeprecated Balanced
Fast Fast 128K 111.6K 16K Snapshot retired (2026-06-29); use gpt-chat-latest instead
gpt-5.2 Flagship
Best Standard 400K 272K 128K Previous flagship, stable and reliable
gpt-5.2-chatDeprecated Balanced
Fast Fast 128K 111.6K 16K Snapshot retired; use gpt-chat-latest instead
gpt-5.2-proPRO Flagship
Best Slowest 400K 272K 128K GPT-5.2 with enhanced reasoning, unlimited compute
gpt-5.1 Flagship
Deep Standard 400K 272K 128K reasoning_effort defaults to none; reasoning must be enabled explicitly
gpt-5.1-chatDeprecated Balanced
Basic Fast 128K 111.6K 16K Snapshot retired; use gpt-chat-latest instead; built-in reasoning, no temperature support
gpt-5 Balanced
Deep Standard 400K 272K 128K Supports reinforcement fine-tuning (RFT, limited rollout)
gpt-5-proPRO Flagship
Best Slowest 400K 272K 128K Enhanced reasoning, more accurate responses
gpt-5-miniRec Balanced
Standard Fast 400K 272K 128K Faster and cheaper, no registration needed
gpt-5-nano Budget
Basic Fastest 400K 272K 128K Fastest, cheapest GPT-5, no registration needed
gpt-5-chatDeprecated Balanced
Fast Fast 128K 111.6K 16K Snapshot retired; use the always-current gpt-chat-latest instead
gpt-4.1 Balanced
Fast Standard 1M 1M 32K Smartest non-reasoning model, fine-tuning supported
gpt-4.1-mini Balanced
Fast Fast 1M 1M 32K Small fast version, fine-tuning supported
gpt-4.1-nano Budget
Fast Fastest 1M 1M 32K Fastest, cheapest GPT-4.1

Codex (Coding)

Optimized for code generation and coding tasks, ideal for agentic workflows, tuned for Codex CLI and the Codex VS Code extension

Model IDTierPriceModalityReasoningSpeedContextInputOutputNotes
gpt-5.3-codexNEW Flagship
Deep Standard 400K 272K 128K Latest Codex model (2026-02), Responses API only
gpt-5.2-codex Flagship
Deep Standard 400K 272K 128K Most capable coding model, optimized for Codex CLI
gpt-5.1-codex Flagship
Deep Standard 400K 272K 128K Responses API only, optimized for Codex CLI
gpt-5.1-codex-max Flagship
Deep Fast 400K 272K 128K Supports reasoning_effort=xhigh
gpt-5.1-codex-mini Balanced
Deep Standard 400K 272K 128K Lightweight Codex
gpt-5-codex Balanced
Deep Standard 400K 272K 128K Responses API only, optimized for Codex CLI
codex-miniDeprecated Balanced
Standard Standard 200K 200K 100K Fine-tuned o4-mini; deprecated, retires 2026-11-15

o-Series Reasoning

Focused on deep thinking and complex reasoning, with configurable reasoning effort reasoning_effort: low/medium/high

Model IDTierPriceModalityReasoningSpeedContextInputOutputNotes
o1-proDeprecated Flagship
Deep Slowest 200K 200K 100K Deprecated, retires 2026-10-21; official recommendation gpt-5.6-sol
o3-deep-researchRec Flagship
Best Slowest 200K 200K 100K Best deep research, extra-long reasoning chains (Azure: via Foundry proxy service only)
o3-proPRO Flagship
Best Slowest 200K 200K 100K More compute, more accurate
o1Deprecated Balanced
Deep Slowest 200K 200K 100K Deprecated, retires 2026-10-21; official recommendation gpt-5.6-sol
o3Deprecated Flagship
Best Slowest 200K 200K 100K Deprecated, retires 2026-10-21; official recommendation gpt-5.6-sol
o4-mini-deep-research Balanced
Deep Standard 200K 200K 100K Faster, cheaper deep research
o4-miniDeprecated Budget
Deep Standard 200K 200K 100K Deprecated, retires 2026-10-16
o3-miniDeprecated Budget
Deep Standard 200K 200K 100K Deprecated, retires 2026-10-01; replacement o4-mini

Image & Video Generation

High-quality visual content from text or images

Model IDTierPriceInputOutputSpeedResolutionNotes
gpt-image-2NEW Flagship
Standard up to 4K Latest-generation image model
gpt-image-1.5 Flagship
Standard up to 4K Most advanced image model
gpt-image-1 Balanced
Slowest up to 2K Previous-generation image model
gpt-image-1-mini Budget
Slowest up to 1K Budget image model
sora-2-pro Flagship
Slowest up to 4K 60fps Legacy video model (officially marked Legacy); still usable
sora-2 Flagship
Slow up to 1080p Legacy video model (officially marked Legacy); still usable

Voice & Audio

Speech-to-text, text-to-speech, realtime chat and audio understanding. gpt-realtime-translate, gpt-realtime-whisper and gpt-live-transcribe are billed by duration; other realtime models by token

Model IDPriceTypeInputOutputSpeedFeaturesNotes
gpt-live-transcribeNEWRec Live transcription
Fastest billed by duration Current recommendation for live transcription (2026-07)
gpt-realtime-whisperNEW Live transcription
Fastest billed by duration Low-latency live transcription (2026-05)
gpt-realtime-translateNEW Realtime translation
Fastest billed by duration Realtime multilingual translation, outputting translated speech and text (2026-05)
gpt-transcribeNEW Speech to text
Standard 25MB files Offline file transcription (/v1/audio/transcriptions), with language hints
whisper-1 Speech to text
Standard 99 languages General speech recognition, supports translation
gpt-4o-transcribe Speech to text
Standard high accuracy GPT-4o-powered transcription
gpt-4o-transcribe-diarize Speech to text
Standard speaker diarization Can distinguish multiple speakers
gpt-realtime-2.1NEWRec Realtime chat
Fast 32K/4K Latest realtime chat (2026-07), improved silence and noise handling
gpt-realtime-2.1-miniNEWPreview Realtime chat
Fastest 32K/4K Lightweight realtime chat
gpt-realtime-2 Realtime chat
Fast 32K/4K Second-generation realtime audio (2026-05)
gpt-realtime-1.5 Realtime chat
Fast 32K/4K Realtime audio processing (2026-02)
gpt-realtimeGA Realtime chat
Fast 32K/4K First GA realtime two-way audio (2025-08)
gpt-realtime-miniDeprecated Realtime chat
Fastest 32K/4K Deprecated, retires 2026-12-15
gpt-audio-1.5NEW Audio understanding
Standard 128K/16K Latest audio understanding model (2026-02)
gpt-audio Audio understanding
Standard 128K/16K Audio support via Chat Completions
gpt-audio-mini Audio understanding
Fastest 128K/16K Lightweight audio processing
tts-1-hdRec Text to speech
Standard high quality High-fidelity text to speech
tts-1 Text to speech
Fast low latency Speed-first TTS
gpt-4o-mini-tts Text to speech
Fast controllable style Can guide the voice with a specific style or tone

Open-Source (Apache 2.0)

Self-hostable, no API call limits, only hardware costs

Model IDPriceParametersModalityReasoningSpeedHardwareContextInputNotes
gpt-oss-120bOSS Self-hosted 120B
Deep Standard 1x H100 80GB 131K 131K Most capable open-source model, requires Foundry deployment
gpt-oss-20bOSS Self-hosted 20B
Deep Standard 1x A100 40GB 131K 131K Supports hosted compute + Foundry Local

Embedding & Tool Models

Text embedding for semantic search, RAG and recommendation systems

Model IDPriceUseInputSpeedDimensionsScoreNotes
text-embedding-3-largeRec Embed
Slow 3,072 MIRACL 54.9 / MTEB 64.6 Most capable embedding model
text-embedding-3-small Embed
Standard 1,536 MIRACL 44.0 / MTEB 62.3 Small embedding model
text-embedding-ada-002 Embed
Slow 1,536 MIRACL 31.4 / MTEB 61.0 Legacy embedding model

Computer-Use Models

Automates computer operation (mouse & keyboard); access requires registration

Model IDPriceModalitySpeedContextInputOutputFeatures
computer-use-previewPreview
Slow 8K 8K 1K Mouse · keyboard · screenshot analysis

GPT-4o Series (Legacy)

Still usable, but GPT-4.1 or GPT-5 is recommended for new projects

Model IDTierPriceModalitySpeedContextInputOutputNotes
gpt-4oDeprecated Balanced
Standard 128K 128K 16K 2024-05/08 versions deprecated (retiring from 2026-10); migrate to gpt-5.1
gpt-4o-miniDeprecated Budget
Fast 128K 128K 16K Deprecated, retires 2027-04-14; migrate to gpt-5-mini
gpt-4-turbo Balanced
Standard 128K 128K 4K Older highly capable model
gpt-4 Balanced
Standard 8K 8K 4K Classic highly capable model
gpt-3.5-turbo Budget
Standard 16K 16K 4K Older GPT with limited capability

Deprecated & Retirement

Retirement info for the models covered on this page. Plan migrations by retirement date and avoid new projects. Lifecycle: Deprecated = announced retirement (still usable, retires on schedule); Retired = no longer available; Legacy = older version (usable, migration advised).

Model IDVersionLifecycleRetirementReplacementMigration advice
gpt-5-chatDeprecated 2025-10-03 Retired 2026-05-13 gpt-chat-latest Snapshot retired; switch to the always-current alias
gpt-5.2-chatDeprecated 2025-12-11 Retired 2026-05-13 gpt-chat-latest Snapshot retired; switch to the always-current alias
gpt-5-chatDeprecated 2025-08-07 Retired 2026-06-29 gpt-chat-latest Snapshot retired; switch to the always-current alias
gpt-5.1-chatDeprecated 2025-11-13 Retired 2026-06-29 gpt-chat-latest Snapshot retired; switch to the always-current alias
gpt-5.2-chatDeprecated 2026-02-10 Retired 2026-06-29 gpt-chat-latest Snapshot retired; switch to the always-current alias
gpt-5.3-chatDeprecated 2026-03-03 Retired 2026-06-29 gpt-chat-latest Snapshot retired; switch to the always-current alias
o3-miniDeprecated 2025-01-31 Deprecated 2026-10-01 o4-mini Drop-in replacement, API-compatible
gpt-4oDeprecated 2024-05-13 Deprecated 2026-10-01 gpt-5.1 Earliest 4o version; migrate early
o4-miniDeprecated 2025-04-16 Deprecated 2026-10-16 No official replacement listed; consider gpt-5.4-mini
sora-2Deprecated 2025-12-08 Legacy 2026-10-15 Still usable; officially marked Legacy, plan migration
sora-2-proDeprecated 2025-10-06 Legacy Still usable; officially marked Legacy, plan migration
o1Deprecated 2024-12-17 Deprecated 2026-10-21 gpt-5.6-sol Official recommendation, stronger reasoning
o1-proDeprecated 2025-03-19 Deprecated 2026-10-21 gpt-5.6-sol Official recommendation, stronger reasoning
o3Deprecated 2025-04-16 Deprecated 2026-10-21 gpt-5.6-sol Official recommendation, stronger reasoning
codex-miniDeprecated 2025-05-16 Deprecated 2026-11-15 Migrate to gpt-5.1-codex-mini
gpt-realtime-miniDeprecated 2025-12-15 Deprecated 2026-12-15 No official replacement listed; consider gpt-realtime-2.1-mini
gpt-4oDeprecated 2024-08-06 Deprecated 2027-04-14 gpt-5.1 Planned migration
gpt-4o-miniDeprecated 2024-07-18 Deprecated 2027-04-14 Migrate to gpt-5-mini or gpt-5.4-mini
gpt-4o 2024-11-20 Legacy 2027-04-14 gpt-5.1 Still usable; GPT-5.1 recommended for new projects
gpt-4.1 2025-04-14 Legacy 2027-04-14 Still usable; GPT-5.4 family recommended for new projects
gpt-4.1-mini 2025-04-14 Legacy 2027-04-14 Still usable; GPT-5.4 family recommended for new projects
gpt-4.1-nano 2025-04-14 Legacy 2027-04-14 Still usable; GPT-5.4 family recommended for new projects

Capability Matrix

Match the best model to your needs quickly

ScenarioRecommendedAlternativesKey capability
Long-document processing~1M tokens gpt-5.6-sol gpt-5.5, gpt-4.1 ~1.05M context
Complex math / logic reasoning o3-deep-research o3-pro, gpt-5.4-pro Best
Agentic coding workflows gpt-5.3-codex gpt-5.2-codex + long tasks
Multimodal understanding (image + audio) gpt-5.6-sol gpt-4o
Realtime voice chat gpt-realtime-2.1 gpt-realtime-2.1-mini WebSocket bidirectional
Live voice transcription gpt-live-transcribe gpt-realtime-whisper billed by duration
High-quality image generation gpt-image-2 gpt-image-1.5 up to 4K resolution
Video generation sora-2 sora-2-pro + synced audio
Ultra-low cost gpt-5.4-nano gpt-5-nano, gpt-4.1-nano Budget
Self-hosted deployment gpt-oss-120b gpt-oss-20b Apache 2.0 open source
Semantic search / RAG text-embedding-3-large text-embedding-3-small 3072-dim vectors
Fastest responses gpt-5.4-nano gpt-4.1-nano Fast