OpenAI models

Text Embedding 3 Small turns text of up to about 8 pages into a fingerprint of its meaning, so search finds passages that mean the same thing even when the words differ. It replaces the older ada-002 and is the sensible default for search over your documents, grouping similar messages or labelling them.

GPT Image 2.5 Sunburst takes a written description, with optional reference pictures, and gives back an image, new or edited, as PNG, JPEG or WebP, with transparent backgrounds on PNG and WebP. It is the precision tier of the 2.5 series, for detailed creative work where an edit has to land exactly as asked, and it adds top quality settings named xhigh and max.

GPT-6 Luna Decisions does not write text. You send content as text, a table of data or a picture, plus up to 200 named questions of three kinds: yes or no, pick among options, or grade against an ordered scale. It returns a probability for every answer, with the likeliest option or expected score, so your own system can set its thresholds for routing, moderation or marking.

Text Embedding 3 Large turns text of up to about 8 pages into a fingerprint of its meaning, so search finds passages that mean the same thing even when the words differ. It is the most accurate of OpenAI's models of this kind, in English and other languages. Choose it when the quality of matches matters most, for search, grouping or labelling documents.

GPT-6 Luna reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the quick, low-cost tier of the GPT-6 series, below GPT-6 Sol, built for heavy traffic and tight response times: chat, sorting messages into categories and light multi-step tasks. Turn its thinking up and it can also take software and computer-use jobs that used to need a Sol-class model.

GPT Image 2.5 Flare takes a written description, with optional reference pictures, and gives back an image, new or edited, as PNG, JPEG or WebP (transparent backgrounds on PNG and WebP). It is the fast tier of the 2.5 series, for everyday volume, creator content and quick drafts, and it adds top quality settings named xhigh and max.

GPT-5.6 Luna reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the fast, lower-cost tier of the GPT-5.6 series, meant for high-volume jobs where speed matters: chat, sorting and labelling messages, and light multi-step tasks. Move up to GPT-5.6 Terra for everyday coding and multi-step work, to GPT-5.6 Sol for command-line and long coding jobs, or to GPT-5.6 Luna Pro when one request needs far deeper thinking.

OpenAI's second GPT Image generation, making and editing high-fidelity pictures from prompts and up to 16 references, in shapes from 21:9 wide to 9:16 tall; the step before the two GPT Image 2.5 tiers

#3 for reasoning · GPT-6.1 Sol reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It upgrades GPT-6 Sol and sits below the flagship Astra, aimed at coding on its own over many steps, operating a computer, professional work built on large documents and multi-step business automation. Compared with the model it replaces, it gets facts wrong less often and follows stated limits and the user's intent more closely, while landing near Astra on these jobs for far less.

GPT-5.4 and GPT Image 2 in one model, so a single conversation can reason, write code and generate or edit pictures without switching models, taking up to 16 reference images

OpenAI's legacy embedding model, kept for systems built on it before text-embedding-3, turning passages of up to about 8K tokens into vectors that measure how closely they relate

GPT Transcribe takes recorded speech in an audio file (mp3, mp4, mpeg, mpga, m4a, wav or webm, up to 25 MB) and gives back a written transcript, either whole or appearing as it is produced. You can add notes about the recording, a list of words you expect to hear, and language hints. OpenAI recommends it for speech transcribed in its original language.

Whisper as OpenAI serves it through its API, the open-source speech recogniser that transcribes and translates audio files up to 25 MB in more than 50 languages, from mp3, mp4, wav or webm

#8 for logos · GPT-5 Image Mini takes text, pictures and files and gives back a picture plus text. It joins GPT-5 Mini's grasp of language to the GPT Image 1 Mini picture-maker, with lettering inside the picture and detailed edits to an existing one. Choose it when you make or edit many pictures from straightforward descriptions.

Whisper Large V3 takes a recording in a common format such as mp3, wav, flac or ogg and gives back a transcript, or a translation into English, in 99 or more languages, with the time of each word and sentence. It gets about one word in ten wrong and copes with noisy recordings and mixed languages.

Whisper Large V3 Turbo takes a recording in a common format such as mp3, wav, flac or ogg and gives back a transcript in 99 or more languages. It is the faster, lighter version of Whisper Large V3 and can turn an hour of audio into text in well under a minute, so it suits large piles of recordings.

#5 for reasoning · This model reads text, pictures and files, taking in about a thousand pages at once, and writes back as many as 120 pages. It thinks before answering, can search the web and run the tools you give it, so it suits everyday coding, writing and multi-step work at the computer. It stands out at making sense of screenshots, and it sits in the middle of its family: the bigger relation above it reads the same amount but costs more to run. It gives back only text, so a job that needs a picture or a voice reply calls for a different kind of model.

The lighter variant of GPT Image 1, for high-quality image generation at reduced latency while keeping transparent backgrounds, quality levels and edits from up to 16 references

GPT-6 Luna Pro is GPT-6 Luna in a Pro mode that thinks much longer before it answers. It reads text, pictures and files and writes back, and each request uses far more thinking, so replies arrive much later. It fits a hard, high-stakes problem where extra accuracy is worth the wait; routine chat, sorting and light multi-step work should stay on plain GPT-6 Luna, and GPT-6 Sol Pro is the step up for heavier coding.

Text Embedding 3 Large turns text of up to about 8 pages into a fingerprint of its meaning, so search finds passages that mean the same thing even when the words differ. It is the most accurate of OpenAI's models of this kind, in English and other languages. Choose it when the quality of matches matters most, for search, grouping or labelling documents.

Text Embedding 3 Small turns text of up to about 8 pages into a fingerprint of its meaning, so search finds passages that mean the same thing even when the words differ. It replaces the older ada-002 and is the sensible default for search over your documents, grouping similar messages or labelling them.

#9 for pictures · GPT-5 Image takes text, pictures and files and gives back a picture with text alongside it. It pairs GPT-5's thinking with GPT Image 1's picture-making, so you can describe a picture, ask for edits to one you upload, and get lettering drawn into the result. Choose it when the request needs thinking before drawing. GPT-5 Image Mini is the lighter version, and GPT-5.4 Image 2 uses the newer image model.

OpenAI's legacy embedding model, kept for systems built on it before text-embedding-3, turning passages of up to about 8K tokens into vectors that measure how closely they relate

#2 for reasoning · GPT-6 Astra reads your text and pictures and writes back, taking in about a thousand pages at once and writing up to 128 pages in one answer. It is the flagship of the GPT-6 series, aimed at big end-to-end jobs such as analysis, building software, researching a subject in depth, science and writing documents, with extra strength on long tasks that drive a computer or browser. GPT-6 Sol is the cost-efficient step below, GPT-6 Luna the fast one, and GPT-6 Astra Pro adds a deeper-thinking mode.

gpt-oss-120b reads text and writes text, keeping about 130 pages in mind at once. OpenAI released it so you can download it and run it yourself, and it is small enough for a single high-end graphics card. You choose how long it thinks (low, medium or high), you can read its full working, and it can use the tools and apps you connect, browse the web and fill in the exact form you give it.

#2 for coding · GPT-5.6 Sol reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the flagship of the GPT-5.6 series, built for hard reasoning, coding and long multi-step tasks, and it stands out on work done in a terminal and on long coding jobs. GPT-5.6 Terra is the balanced tier below it, GPT-5.6 Luna the fast one, and GPT-5.6 Sol Pro runs the same model with deeper thinking for the hardest requests.

#9 for reasoning · GPT-6 Sol reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the lower-cost high-end model of the GPT-6 series, between the flagship Astra and the fast Luna, suited to heavy professional work, automating business workflows and computer use, and strongest on long software tasks inside existing projects.

GPT-4o mini reads text and pictures and writes text back, keeping about 128 pages in mind and replying up to 16 pages, and it can use connected tools and fill in exact forms. It is the small member of the GPT-4o family and the one OpenAI recommends in place of GPT-3.5 Turbo. Choose it for high-volume chat, pulling details out of documents or tagging pictures where GPT-4o's extra ability is not needed. GPT-4.1 mini reads about a thousand pages if your inputs are longer.

#7 for photo editing · OpenAI's first GPT Image model, accurate at writing text into pictures, able to return transparent backgrounds and to edit from up to 16 reference images, making up to ten images a request at four quality levels

#5 for coding · GPT-5.6 Terra reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the balanced tier of the GPT-5.6 series, sitting between the flagship Sol and the lighter Luna, and suits everyday coding, reasoning and multi-step work when you want strong results without paying for the top tier.

gpt-oss-20b reads text and writes text, keeping about 130 pages in mind at once. It is the smaller of OpenAI's two downloadable models, free to run yourself under an open license, quick enough to answer fast and light enough for an ordinary computer or a single graphics card. You choose how long it thinks, it can use the tools and apps you connect, and it can be retrained on your own examples.

GPT-5 Mini reads text and pictures and writes text back, keeping about 400 pages in mind, and it thinks before answering and can use connected tools. It is a scaled-down GPT-5 for lighter thinking jobs, keeping GPT-5's instruction handling and safety training at quicker speed, and it succeeds o4-mini. Pick it for chat features, pulling out details and routine coding help that still need some thought.

GPT-4.1 Mini reads text and pictures and writes text back, taking in the same thousand pages as GPT-4.1, with replies up to 32 pages. It is the mid-sized model of the series, built to match GPT-4o with quicker answers, so it suits interactive apps such as chat assistants and coding helpers that reply while the user waits.

GPT-5.4 nano reads text and pictures and writes text back, keeping about 400 pages in mind. It is the lightest GPT-5.4 model, built for speed on sorting items into classes, pulling out fields, ranking results and small helper steps that run in the background or in real time, and it favours quick replies over deep thinking.

GPT-5.6 Luna Pro is GPT-5.6 Luna in a Pro mode that thinks much longer before it answers. It reads the same text, pictures and files and writes back, still taking in about a thousand pages at once, but each request uses far more thinking, so answers arrive much later. GPT-5.6 Sol Pro gives the flagship the same mode.

A faster, more efficient GPT-5.4 for production at scale, serving chat apps, coding assistants and agent workflows with reliable instruction following and solid multi-step reasoning over text and images

GPT-6 Luna reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the quick, low-cost tier of the GPT-6 series, below GPT-6 Sol, built for heavy traffic and tight response times: chat, sorting messages into categories and light multi-step tasks. Turn its thinking up and it can also take software and computer-use jobs that used to need a Sol-class model.

GPT-5.4 reads text and pictures and writes text back, taking in about 900 pages at once and writing up to 128 pages in one answer. It folds OpenAI's Codex models and its GPT models into one, so a single request can write code, read documents, use connected tools and follow detailed instructions with fewer retries. Choose it as the one model for mixed coding and general work. GPT-5.4 mini and nano serve high volume, GPT-5.4 Pro is for hard problems, and GPT-5.5 is the later release.

GPT-6.1 Sol Pro is the same underlying model as GPT-6.1 Sol, served with `reasoning.mode` set to `pro` for higher-quality responses on complex tasks. **Cost note:** pro mode spends far more...

GPT-4.1 reads text and pictures and writes text back, taking in about a thousand pages at once, with replies up to 32 pages. It is tuned to follow instructions exactly, write precise code changes and find facts in very long documents, so use it for assistants that act on their own, coding tools and answering from a large set of documents. Its knowledge runs to June 2024.

GPT-5.2 reads text, pictures and files and writes text back, keeping about 400 pages in mind, with replies up to 128 pages. Choose it as the general model of that generation. GPT-5.2 Chat is the quick conversational version, GPT-5.2 Pro is for high-stakes thinking, and GPT-5.2-Codex is the coding specialist.

This model reads text, pictures and files, taking in about a thousand pages at once, and writes back as many as 120 pages of text. It is the same brain as its smaller relation but run in a deeper thinking mode, so it takes longer and spends more effort on each answer to handle the hardest jobs well. It thinks before answering, searches the web, and runs the tools you give it, which suits complex analysis, careful long-form writing and multi-step work built on large documents. It gives back only text, so a job that needs a picture or a voice reply calls for a different kind of model.

GPT-5 Nano reads text and pictures and writes text back, keeping about 400 pages in mind. It is the fastest and smallest model of the GPT-5 family, made for developer tools and real-time interactions, and it succeeds GPT-4.1 nano. Its thinking is shallower than the larger models'.

#6 for slide decks · GPT-5.5 reads your text, pictures and files and writes back, taking in about 900 pages at once and writing up to 128 pages in one answer. Over its predecessor it brings sharper thinking, fewer failures and less wasted effort on difficult work, and it is aimed at complex professional jobs that mix coding and images.

GPT-5.1 reads text, pictures and files and writes text back, keeping about 400 pages in mind, with replies up to 128 pages. It adjusts its thinking to the question, answering simple ones fast and spending more effort on hard ones, and its explanations use less jargon than GPT-5's. Choose it as a general model for coding, math and analysis with dependable use of connected tools. GPT-5.2 adds stronger long-task and long-document performance; GPT-5.1-Codex narrows the model to coding.

The same GPT-5.6 Sol flagship run in pro reasoning mode, taking more care over complex command-line work, long coding tasks and hard problems to give higher-quality answers, over text, images and files

The hardest problems on OpenAI's flagship, where extra reasoning work per answer is worth the latency and tokens

GPT-5.6 Luna reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the fast, lower-cost tier of the GPT-5.6 series, meant for high-volume jobs where speed matters: chat, sorting and labelling messages, and light multi-step tasks. Move up to GPT-5.6 Terra for everyday coding and multi-step work, to GPT-5.6 Sol for command-line and long coding jobs, or to GPT-5.6 Luna Pro when one request needs far deeper thinking.

GPT-5.3-Codex reads text, pictures and files and writes text back, keeping about 400 pages in mind. It joins GPT-5.2-Codex's software engineering with GPT-5.2's broader thinking, so it handles long work with connected tools such as debugging, deployment and terminal tasks, and also knowledge work like drafting documents, analysing spreadsheets and building slides. You can steer it while it runs. GPT-5.2-Codex is the code-only predecessor; GPT-5.4 merges the Codex and GPT lines.

The fastest and smallest GPT-4.1, keeping a million-token context and scoring 80.1% on MMLU, above GPT-4o mini, for low-latency jobs such as classification and autocomplete

GPT-4o reads text and pictures and writes text back, including answers that follow an exact form you give it, keeping about 128 pages in mind with replies up to 16 pages. Its knowledge stops in October 2023, and it handles non-English text and images better than earlier models. Choose it for an assistant that reads screenshots or photos and answers in many languages.

GPT-4o-mini (2024-07-18) is the dated launch release of GPT-4o mini. It reads text and pictures, writes text back, keeps about 128 pages in mind and replies up to 16 pages, and it can use connected tools and fill in exact forms. Pick it when a pipeline must stay pinned to that release so results do not shift.

GPT-5 reads text and pictures and writes text back, keeping about 400 pages in mind, with replies up to 128 pages. It thinks before it answers, with four effort levels from minimal to high, and is tuned for step-by-step problems, exact instruction following and fewer invented answers. Its knowledge runs to September 2024.

#3 for reasoning · GPT-6.1 Sol reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It upgrades GPT-6 Sol and sits below the flagship Astra, aimed at coding on its own over many steps, operating a computer, professional work built on large documents and multi-step business automation. Compared with the model it replaces, it gets facts wrong less often and follows stated limits and the user's intent more closely, while landing near Astra on these jobs for far less.

GPT-6 Sol Pro is GPT-6 Sol in a Pro mode. It reads text, pictures and files and writes back, but each request uses far more thinking and takes much longer. Keep it for the hardest software and workflow problems, where accuracy outweighs waiting. Everyday multi-step and coding work stays on plain GPT-6 Sol, and GPT-6.1 Sol Pro is the same mode on the upgraded Sol.

This model reads text, pictures and files, taking in about a thousand pages at once, and writes back as many as 120 pages of text. It is the same brain as its smaller relation but run in a deeper thinking mode, so it takes longer and spends more effort on each answer to handle the hardest jobs well. It thinks before answering, searches the web, and runs the tools you give it, which suits complex analysis, careful long-form writing and multi-step work built on large documents. It gives back only text, so a job that needs a picture or a voice reply calls for a different kind of model.

GPT-5.1-Codex-Mini reads text and pictures and writes text back, keeping about 400 pages in mind, and it thinks before answering and can use connected tools. It is a smaller, quicker sibling of the Codex model, meant for the same kind of coding-assistant work on a lighter footprint.

gpt-oss-safeguard-20b is built for moderation, not general chat. You send your own written policy first and the content to check second; it thinks through the content against that policy and returns a verdict in whatever format you ask for, from a 0 or 1 to a short explanation. Changing the policy needs no retraining.

GPT-5.6 Terra Pro is GPT-5.6 Terra in a Pro mode. It reads text, pictures and files and writes back, but each request uses far more thinking and takes much longer. It is for hard problems where accuracy justifies the wait; plain GPT-5.6 Terra handles the everyday coding and multi-step work, while GPT-5.6 Sol Pro takes the longest coding jobs and GPT-5.6 Luna Pro the lightest ones.

GPT-6 Luna Pro is GPT-6 Luna in a Pro mode that thinks much longer before it answers. It reads text, pictures and files and writes back, and each request uses far more thinking, so replies arrive much later. It fits a hard, high-stakes problem where extra accuracy is worth the wait; routine chat, sorting and light multi-step work should stay on plain GPT-6 Luna, and GPT-6 Sol Pro is the step up for heavier coding.

GPT-4o (2024-08-06) reads text and pictures and writes text back, keeping about 128 pages in mind, with replies up to 16 pages. Its change from the May release is that you can hand it an exact form and get an answer that follows it. Pick it when a pipeline needs form-shaped answers from a model that reads pictures. The 2024-11-20 version writes better prose and works through uploaded files more thoroughly.

#5 for reasoning · This model reads text, pictures and files, taking in about a thousand pages at once, and writes back as many as 120 pages. It thinks before answering, can search the web and run the tools you give it, so it suits everyday coding, writing and multi-step work at the computer. It stands out at making sense of screenshots, and it sits in the middle of its family: the bigger relation above it reads the same amount but costs more to run. It gives back only text, so a job that needs a picture or a voice reply calls for a different kind of model.

A name that always points at the newest Instant chat model used in ChatGPT and moves to each update automatically, for apps that want ChatGPT's current behaviour, reading text, images and files with tools

gpt-oss-20b reads text and writes text, keeping about 130 pages in mind at once. It is the smaller of OpenAI's two downloadable models, free to run yourself under an open license, quick enough to answer fast and light enough for an ordinary computer or a single graphics card. You choose how long it thinks, it can use the tools and apps you connect, and it can be retrained on your own examples.

GPT-4o (2024-11-20) reads text and pictures and writes text back, keeping about 128 pages in mind, with replies up to 16 pages. This version improved creative writing, with more natural and better-aimed prose, and it gives deeper, more thorough answers about uploaded files. Pick it for drafting and for questions about attached documents. The May and August 2024 versions suit holding results to that exact release.

OpenAI's o3, a well-rounded reasoning model that set a new standard in maths, science, coding and visual reasoning, and is strong at technical writing, for problems that need analysis across text, code and images

GPT-5.1-Codex reads text and pictures, including screenshots of app screens, and writes text back. It is GPT-5.1 specialised for software engineering: building projects from scratch, adding features, debugging, large restructurings and code review, in short sessions or long independent runs, with thinking effort you can set. It sticks closer to a developer's instructions than GPT-5.1.

gpt-oss-120b reads text and writes text, keeping about 130 pages in mind at once. OpenAI released it so you can download it and run it yourself, and it is small enough for a single high-end graphics card. You choose how long it thinks (low, medium or high), you can read its full working, and it can use the tools and apps you connect, browse the web and fill in the exact form you give it.

GPT-5.2-Codex reads text and pictures and writes text back, keeping about 400 pages in mind. It is a newer release of the 5.1 Codex model for software engineering: new projects, features, debugging, large restructurings and code review, in interactive sessions or long independent runs, with adjustable thinking effort. It follows developer instructions more closely and writes cleaner code.

GPT-5 Nano reads text and pictures and writes text back, keeping about 400 pages in mind. It is the fastest and smallest model of the GPT-5 family, made for developer tools and real-time interactions, and it succeeds GPT-4.1 nano. Its thinking is shallower than the larger models'.

#5 for coding · GPT-5.6 Terra reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the balanced tier of the GPT-5.6 series, sitting between the flagship Sol and the lighter Luna, and suits everyday coding, reasoning and multi-step work when you want strong results without paying for the top tier.

#9 for reasoning · GPT-6 Sol reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the lower-cost high-end model of the GPT-6 series, between the flagship Astra and the fast Luna, suited to heavy professional work, automating business workflows and computer use, and strongest on long software tasks inside existing projects.

GPT-3.5 Turbo (older v0613) is a frozen, dated version of GPT-3.5 Turbo that reads text and writes text back, keeping only about 4 pages in mind. It can use connected tools and fill in exact forms, and its knowledge stops in September 2021. Use it only to reproduce results from that exact version.

GPT-5.6 Luna Pro is GPT-5.6 Luna in a Pro mode that thinks much longer before it answers. It reads the same text, pictures and files and writes back, still taking in about a thousand pages at once, but each request uses far more thinking, so answers arrive much later. GPT-5.6 Sol Pro gives the flagship the same mode.

GPT-5.4 reads text and pictures and writes text back, taking in about 900 pages at once and writing up to 128 pages in one answer. It folds OpenAI's Codex models and its GPT models into one, so a single request can write code, read documents, use connected tools and follow detailed instructions with fewer retries. Choose it as the one model for mixed coding and general work. GPT-5.4 mini and nano serve high volume, GPT-5.4 Pro is for hard problems, and GPT-5.5 is the later release.

GPT-5.1-Codex-Max reads text and pictures and writes text back, keeping about 400 pages in mind. It is the Codex model for software jobs that run long and hold a lot in mind, built on an updated 5.1 thinking base and trained on work in engineering, mathematics and research that it carries out step by step on its own. Use it for an assistant that works through a large codebase for hours. GPT-5.1-Codex is the standard one, and GPT-5.1-Codex-Mini the smaller.

GPT-4o (2024-05-13) is the May 2024 dated release of GPT-4o: it reads text and pictures, writes text back and keeps about 128 pages in mind, but its replies stop at 4 pages. Pick it only to reproduce results from that release. The 2024-08-06 version can follow an exact form you give it and replies up to 16 pages, and the 2024-11-20 version writes more naturally.

#2 for reasoning · GPT-6 Astra reads your text and pictures and writes back, taking in about a thousand pages at once and writing up to 128 pages in one answer. It is the flagship of the GPT-6 series, aimed at big end-to-end jobs such as analysis, building software, researching a subject in depth, science and writing documents, with extra strength on long tasks that drive a computer or browser. GPT-6 Sol is the cost-efficient step below, GPT-6 Luna the fast one, and GPT-6 Astra Pro adds a deeper-thinking mode.

GPT-5.5 Pro reads your text, pictures and files and writes back, taking in about 900 pages at once and writing up to 128 pages in one answer. It is the deep-thinking tier of GPT-5.5, made for accuracy when mistakes are expensive: problems that run over many steps, coding done step by step on its own, and exact follow-through.

GPT-4 reads text and writes text back, keeping about 8 pages in mind, with replies up to 4 pages, and it can use connected tools and fill in exact forms. Its knowledge stops in September 2021, and it is the original GPT-4 generation. Choose it only to keep an older setup steady. GPT-4 Turbo reads 128 pages and accepts pictures, and GPT-4.1 reads about a thousand pages with a later knowledge cutoff.

A faster, more efficient GPT-5.4 for production at scale, serving chat apps, coding assistants and agent workflows with reliable instruction following and solid multi-step reasoning over text and images

The hardest problems on OpenAI's flagship, where extra reasoning work per answer is worth the latency and tokens

The long-context GPT-3.5 Turbo, with four times the original window, about 20 pages in a single request, for chat and completions on longer documents, with tool calling and data up to September 2021

#2 for coding · GPT-5.6 Sol reads your text, pictures and files and writes back, and it can take in about a thousand pages at once. It is the flagship of the GPT-5.6 series, built for hard reasoning, coding and long multi-step tasks, and it stands out on work done in a terminal and on long coding jobs. GPT-5.6 Terra is the balanced tier below it, GPT-5.6 Luna the fast one, and GPT-5.6 Sol Pro runs the same model with deeper thinking for the hardest requests.

GPT-5.4 Pro reads your text, pictures and files and writes back, taking in about 900 pages at once and writing up to 128 pages in one answer. It is the Pro version of GPT-5.4, tuned to think through a task stage by stage, follow instructions exactly and stay accurate on high-stakes work such as coding on its own over many steps or a long, tangled problem.

The Pro version of GPT-5, above the standard model, for complex, high-stakes tasks that need step-by-step reasoning and accuracy, with fewer hallucinations, less sycophancy and stronger coding, writing and health answers

GPT-6.1 Sol Pro is the same underlying model as GPT-6.1 Sol, served with `reasoning.mode` set to `pro` for higher-quality responses on complex tasks. **Cost note:** pro mode spends far more...

o3-pro reads text and pictures and writes text, keeping about 200 pages in mind at once and writing up to 100 pages in one answer. It is o3 given more computing power to think, aimed at difficult problems where consistently better answers are worth the extra time and spend. Use plain o3 for ordinary multi-step reasoning, and o1-pro for the extra-effort version of o1.

The Pro version of GPT-5.2, above the standard model, improving on GPT-5 Pro in agentic coding and long contexts for complex, high-stakes tasks that need step-by-step reasoning and accuracy

GPT-4o mini reads text and pictures and writes text back, keeping about 128 pages in mind and replying up to 16 pages, and it can use connected tools and fill in exact forms. It is the small member of the GPT-4o family and the one OpenAI recommends in place of GPT-3.5 Turbo. Choose it for high-volume chat, pulling details out of documents or tagging pictures where GPT-4o's extra ability is not needed. GPT-4.1 mini reads about a thousand pages if your inputs are longer.

GPT-5.4 nano reads text and pictures and writes text back, keeping about 400 pages in mind. It is the lightest GPT-5.4 model, built for speed on sorting items into classes, pulling out fields, ranking results and small helper steps that run in the background or in real time, and it favours quick replies over deep thinking.

GPT-6 Sol Pro is GPT-6 Sol in a Pro mode. It reads text, pictures and files and writes back, but each request uses far more thinking and takes much longer. Keep it for the hardest software and workflow problems, where accuracy outweighs waiting. Everyday multi-step and coding work stays on plain GPT-6 Sol, and GPT-6.1 Sol Pro is the same mode on the upgraded Sol.

GPT-5.6 Terra Pro is GPT-5.6 Terra in a Pro mode. It reads text, pictures and files and writes back, but each request uses far more thinking and takes much longer. It is for hard problems where accuracy justifies the wait; plain GPT-5.6 Terra handles the everyday coding and multi-step work, while GPT-5.6 Sol Pro takes the longest coding jobs and GPT-5.6 Luna Pro the lightest ones.

The same GPT-5.6 Sol flagship run in pro reasoning mode, taking more care over complex command-line work, long coding tasks and hard problems to give higher-quality answers, over text, images and files

GPT-4 Turbo reads text and pictures and writes text back, keeping about 128 pages in mind, with replies up to 4 pages. A request with pictures can also ask for an exact form as the answer and use connected tools, and its knowledge runs to December 2023. Pick it to keep an older picture-and-tools setup unchanged. For new picture-reading work, GPT-4o keeps the same 128 pages with replies up to 16 pages, and GPT-4 reads only 8 pages and no pictures.

GPT-4.1 reads text and pictures and writes text back, taking in about a thousand pages at once, with replies up to 32 pages. It is tuned to follow instructions exactly, write precise code changes and find facts in very long documents, so use it for assistants that act on their own, coding tools and answering from a large set of documents. Its knowledge runs to June 2024.

GPT-4.1 Mini reads text and pictures and writes text back, taking in the same thousand pages as GPT-4.1, with replies up to 32 pages. It is the mid-sized model of the series, built to match GPT-4o with quicker answers, so it suits interactive apps such as chat assistants and coding helpers that reply while the user waits.

The fastest and smallest GPT-4.1, keeping a million-token context and scoring 80.1% on MMLU, above GPT-4o mini, for low-latency jobs such as classification and autocomplete

GPT-4o reads text and pictures and writes text back, including answers that follow an exact form you give it, keeping about 128 pages in mind with replies up to 16 pages. Its knowledge stops in October 2023, and it handles non-English text and images better than earlier models. Choose it for an assistant that reads screenshots or photos and answers in many languages.

GPT-5 reads text and pictures and writes text back, keeping about 400 pages in mind, with replies up to 128 pages. It thinks before it answers, with four effort levels from minimal to high, and is tuned for step-by-step problems, exact instruction following and fewer invented answers. Its knowledge runs to September 2024.

GPT-5 Mini reads text and pictures and writes text back, keeping about 400 pages in mind, and it thinks before answering and can use connected tools. It is a scaled-down GPT-5 for lighter thinking jobs, keeping GPT-5's instruction handling and safety training at quicker speed, and it succeeds o4-mini. Pick it for chat features, pulling out details and routine coding help that still need some thought.

The Pro version of GPT-5, above the standard model, for complex, high-stakes tasks that need step-by-step reasoning and accuracy, with fewer hallucinations, less sycophancy and stronger coding, writing and health answers

GPT-5.1 reads text, pictures and files and writes text back, keeping about 400 pages in mind, with replies up to 128 pages. It adjusts its thinking to the question, answering simple ones fast and spending more effort on hard ones, and its explanations use less jargon than GPT-5's. Choose it as a general model for coding, math and analysis with dependable use of connected tools. GPT-5.2 adds stronger long-task and long-document performance; GPT-5.1-Codex narrows the model to coding.

GPT-5.2 reads text, pictures and files and writes text back, keeping about 400 pages in mind, with replies up to 128 pages. Choose it as the general model of that generation. GPT-5.2 Chat is the quick conversational version, GPT-5.2 Pro is for high-stakes thinking, and GPT-5.2-Codex is the coding specialist.

GPT-5.2 Chat, also called Instant, reads text, pictures and files and writes text back, keeping about 128 pages in mind, with replies up to 32 pages. It is tuned for quick conversation: it answers short questions fast, thinks only on harder ones, and keeps a warmer tone with steadier thinking on short questions. Use it for a chatbot or assistant that must feel responsive.

The Pro version of GPT-5.2, above the standard model, improving on GPT-5 Pro in agentic coding and long contexts for complex, high-stakes tasks that need step-by-step reasoning and accuracy

GPT-5.4 Pro reads your text, pictures and files and writes back, taking in about 900 pages at once and writing up to 128 pages in one answer. It is the Pro version of GPT-5.4, tuned to think through a task stage by stage, follow instructions exactly and stay accurate on high-stakes work such as coding on its own over many steps or a long, tangled problem.

#6 for slide decks · GPT-5.5 reads your text, pictures and files and writes back, taking in about 900 pages at once and writing up to 128 pages in one answer. Over its predecessor it brings sharper thinking, fewer failures and less wasted effort on difficult work, and it is aimed at complex professional jobs that mix coding and images.

GPT-5.5 Pro reads your text, pictures and files and writes back, taking in about 900 pages at once and writing up to 128 pages in one answer. It is the deep-thinking tier of GPT-5.5, made for accuracy when mistakes are expensive: problems that run over many steps, coding done step by step on its own, and exact follow-through.

OpenAI's o3, a well-rounded reasoning model that set a new standard in maths, science, coding and visual reasoning, and is strong at technical writing, for problems that need analysis across text, code and images

o3-mini reads text and files, not pictures, and writes text, keeping about 200 pages in mind at once. It is the economical thinking model of the o3 line, built for science, math and coding, with an effort level you set to low, medium or high. At medium effort it matches o1 on hard math and science tests, and it can call functions and fill in the exact form you give it.

o4-mini reads text, pictures and files and writes back, keeping about 200 pages in mind at once. It is the small thinking model of its line: accurate on math and science, able to solve visual problems, edit code, use connected tools and fill in the exact form you give it, and quick enough for steady heavy traffic. It beats o3-mini and gets close to o3 in some areas.