If you have opened ChatGPT recently and wondered which model you are actually talking to, you are not alone. OpenAI now ships several models at once, retires old ones on a rolling schedule, and just changed the way it names them. This guide clears it up in plain language: the current version, what changed with the new GPT-5.6 launch, the full history from GPT-1 to today, and a simple way to pick the right model for what you are doing.
Everything below reflects OpenAI's lineup as of July 10, 2026, one day after the GPT-5.6 general release.
What is the latest version of ChatGPT right now?
The latest version of ChatGPT is the GPT-5.6 family, made up of three models named Sol, Terra, and Luna. OpenAI moved this family from a limited preview to general availability on July 9, 2026, across ChatGPT, ChatGPT Work, Codex, and the OpenAI API.
There is one detail that trips most people up, so it is worth being precise about it:
The newest model family is GPT-5.6 (Sol, Terra, Luna).
The default model you chat with in ChatGPT is still GPT-5.5 Instant, which has been the standard for every plan since May 5, 2026.
GPT-5.6 Sol powers the deeper reasoning modes for paid users, and GPT-5.6 Sol Pro powers the top Pro option.
In short: if you type a normal question into ChatGPT today, GPT-5.5 Instant usually answers it. When you switch on reasoning on a paid plan, GPT-5.6 Sol takes over. Both are current. They just do different jobs.
Key takeaways
The latest ChatGPT release is GPT-5.6, split into Sol (flagship), Terra (balanced), and Luna (fast and cheap), live since July 9, 2026.
GPT-5.5 Instant remains the everyday default across all plans, including Free.
Free users do not get GPT-5.6 in normal chat. Paid plans unlock GPT-5.6 Sol reasoning, and Free or Go users can only reach Terra inside ChatGPT Work and Codex.
OpenAI introduced a new naming system: the number is the generation, and Sol, Terra, and Luna are capability tiers that carry forward across future versions.
On OpenAI's own benchmark tables, Sol leads several agentic and coding-agent tests, while Claude models still lead on some coding evaluations such as SWE-Bench Pro. Treat all vendor numbers as a starting point, not the final word.
The GPT-5.6 launch: what actually changed
OpenAI first previewed GPT-5.6 Sol on June 26, 2026 for a small group of trusted partners, a restricted start tied to an ongoing review process with the U.S. government. The full public rollout followed on July 9, 2026. Here is what matters for real use.
Sol, Terra, and Luna: three models, one generation
Instead of releasing a single model, OpenAI shipped a family with clearly different jobs:
GPT-5.6 Sol is the flagship. It is built for the hardest work: complex coding, research, science, cybersecurity, computer use, and design. On the Artificial Analysis Coding Agent Index, OpenAI reports Sol setting a new high of 80.0, ahead of Claude Fable 5, while using fewer output tokens and finishing faster. There is also a Sol Pro option in ChatGPT for the most demanding tasks and longer running workflows.
GPT-5.6 Terra is the balanced everyday tier. It aims to deliver most of Sol's usefulness at a much lower cost, which makes it the practical default for teams that want to control their spend without a big quality drop.
GPT-5.6 Luna is the fastest and cheapest of the three. It is designed for high volume work where speed and price matter more than squeezing out the last bit of accuracy. OpenAI also plans a Cerebras deployment for Sol running at up to 750 tokens per second in July.
A new naming system worth understanding
GPT-5.6 introduced a naming change that will shape every future release. The number (5.6) marks the generation. The names (Sol, Terra, Luna) mark durable capability tiers that can improve on their own schedule. So when a future generation arrives, expect the tier names to carry over. Once you learn the pattern once, you will not have to relearn it every few months.
GPT-5.6 API pricing
For developers, OpenAI prices the three tiers per one million tokens as follows:
Model | Input | Output |
|---|---|---|
GPT-5.6 Sol | $5.00 | $30.00 |
GPT-5.6 Terra | $2.50 | $15.00 |
GPT-5.6 Luna | $1.00 | $6.00 |
In the API, the bare gpt-5.6 alias routes to gpt-5.6-sol. If you want predictable routing, call the explicit IDs: gpt-5.6-sol, gpt-5.6-terra, or gpt-5.6-luna. Prompt caching also changed: from GPT-5.6 onward, cache writes are billed at 1.25 times the uncached input rate, cache reads keep the 90 percent discount, and there is a 30 minute minimum cache life with explicit cache breakpoints.
New capabilities: max, ultra, and Programmatic Tool Calling
Three additions stand out at launch:
Max reasoning effort adds a higher tier of thinking for genuinely hard problems. It is available to users with GPT-5.6 access in ChatGPT Work and Codex.
Ultra mode runs four agents in parallel to attack a complex task from several angles at once. OpenAI reports this lifts the Terminal-Bench 2.1 score from 88.8 percent to 91.9 percent. Ultra is available to Pro and Enterprise users in ChatGPT Work, and to Plus and higher plans in Codex.
Programmatic Tool Calling lets the model write and run its own JavaScript inside an isolated runtime with no network access, which is useful for calculation and data shaping steps without wiring up a custom tool for every case.
How to access GPT-5.6 by plan
This is where "the latest version" gets specific to you:
ChatGPT (standard chat): Plus, Pro, Business, and Enterprise users get GPT-5.6 Sol at medium and higher reasoning effort. Pro and Enterprise can also select Sol Pro. Terra and Luna are not selectable here.
Free and logged out users: No GPT-5.6 in standard chat. Your default stays GPT-5.5 Instant.
ChatGPT Work and Codex: Free and Go users get Terra. Plus and higher plans can choose Sol, Terra, or Luna and set the effort per model.
OpenAI API: All three tiers are self serve for any account.
Availability is rolling out gradually, so if GPT-5.6 has not appeared in your picker yet, it may still reach your account over the following days.
The current ChatGPT model picker, explained
OpenAI spent 2026 simplifying a messy picker. Where 2025 had GPT-4o, GPT-4.1, o1, o3, o3-pro, and o4-mini all fighting for space, most people now see a short list. Here is what each choice actually runs today.
Instant
Instant is the default for fast, everyday answers. It runs GPT-5.5 Instant on every plan, including Free. It handles writing, summaries, translation, quick research, and most coding questions without slowing down to reason. If you change nothing, this is what you get.
Thinking
Thinking is the reasoning path for harder problems: tricky math, multi step logic, detailed research, spreadsheet building, and careful code. On paid plans this now taps GPT-5.6 Sol at medium or higher effort, or GPT-5.5 Thinking. Free and Go users can reach a lighter GPT-5.4 mini reasoning option through the Thinking toggle in the plus menu, but not full Sol.
Pro
Pro pushes reasoning to its ceiling for the most demanding work, such as legal analysis, financial modeling, and research grade problems. It runs GPT-5.5 Pro, and on Pro and Enterprise plans, GPT-5.6 Sol Pro. This tier trades speed for depth and reliability.
Auto
Auto lets ChatGPT read your prompt and route it to the right model, sending simple questions to a fast model and harder ones to reasoning. It is a good default if you do not want to think about model selection at all.
Which ChatGPT model should you use?
You do not need to memorize version numbers. Match the model to the job.
Everyday writing, email, summaries, quick questions: Instant (GPT-5.5 Instant). It is fast and accurate enough for the large majority of tasks.
Hard math, logic, or careful multi step reasoning: Thinking, which reaches GPT-5.6 Sol on paid plans.
Serious coding and repository work: Codex with a GPT-5.6 tier, or Thinking for smaller code tasks. For long agentic sessions, Sol or Sol Pro earn their higher cost.
Spreadsheets, slide decks, and document heavy work: Thinking or Pro, which handle structured output and layout more reliably than Instant.
Legal, financial, medical, or research grade analysis: Pro (GPT-5.5 Pro or Sol Pro) for the extra accuracy and lower error rate.
High volume, cost sensitive automation through the API: Terra for balance, or Luna when speed and price come first.
A simple rule: start on Instant, switch to Thinking the moment an answer feels shallow or a task has real stakes, and reserve Pro for the problems where being wrong is expensive.
Every ChatGPT version, from GPT-1 to GPT-5.6
If you want the full lineage, here is how OpenAI got from a research proof of concept to today's family. This is the "chatgpt models" history in order.
GPT-1 (June 2018)
The first GPT model, with 117 million parameters. It could produce simple text and mainly proved that the Transformer architecture could learn language patterns from books. Practical uses were limited, but the direction was set.
GPT-2 (February 2019)
Ten times larger at 1.5 billion parameters, trained on millions of web pages. OpenAI initially held back the full model over misuse concerns, which kicked off a lasting debate about responsible release. Text quality and coherence jumped noticeably.
GPT-3 (June 2020)
The model that made the wider world pay attention. At 175 billion parameters, it could write human sounding text and even generate code, and it handled new tasks from just a few examples. Offered commercially through an API, it seeded a first wave of AI products.
GPT-3.5 and the ChatGPT launch (November 2022)
Strictly a refinement of GPT-3, but trained with human feedback to hold a conversation and admit mistakes. Packaged as ChatGPT, it reached a million users in five days and turned generative AI into a mainstream tool overnight.
GPT-4 (March 2023)
A large step up in reasoning and accuracy, and the first mainstream OpenAI model that could interpret images. It processed longer context and performed strongly on standardized exams across law, medicine, and programming.
GPT-4 Turbo (November 2023)
A faster, cheaper GPT-4 with a 128,000 token context window, JSON mode for structured output, and more current training data. It made advanced capability affordable enough for far more applications.
GPT-4o (May 2024)
The "omni" model, built to handle text, images, and audio together with near real time responses. It brought voice conversation and multimodal understanding to a much wider audience, with a lighter GPT-4o mini for cost sensitive uses.
o1 and the reasoning era (late 2024)
OpenAI's first models designed to "think" before answering, aimed at complex problem solving in science, math, and coding. The o1 line paved the way for reasoning to become a standard feature rather than a niche mode.
o3, o4-mini, and GPT-4.1 (2025)
The o3 and o4 series pushed reasoning further and, for the first time, let models use every ChatGPT tool together to complete tasks. GPT-4.1, released in April 2025, brought a 1 million token context window and much better instruction following for long documents and code.
GPT-4.5 (February 2025)
A general model praised more for the natural feel of its writing and conversation than for record benchmarks. It shipped first as a research preview, then rolled out to paying tiers. It was retired from ChatGPT on June 26, 2026.
GPT-5 (August 2025)
A milestone: OpenAI merged reasoning and standard responses under one interface, and gave free users access to a reasoning model for the first time. The family included GPT-5, plus mini and nano variants and a Pro version, with configurable reasoning effort and a big drop in factual errors when web search was on.
GPT-5.1 (November 2025)
A refresh focused on experience rather than raw scores. It brought adaptive reasoning, so the model decided when to think, a warmer default tone, and account level personality settings for tone, humor, formality, and response length. It was retired from ChatGPT on March 11, 2026.
GPT-5.2 (late 2025 to early 2026)
An efficiency focused update that improved everyday responses and how-to answers, and served as the bridge to the faster GPT-5.3 and GPT-5.4 releases. It was retired from ChatGPT on June 12, 2026.
GPT-5.3 (February 2026)
Centered on speed and a strong agentic coding model, GPT-5.3 Codex, which combined the Codex and GPT-5 training stacks. GPT-5.3 Instant became the default for a stretch of early 2026 before GPT-5.5 Instant took over.
GPT-5.4 (March 2026)
A single frontier model that folded in GPT-5.3 Codex coding strength while improving work across spreadsheets, presentations, and documents. It added native computer use and a Thinking mode that can show its plan mid response. GPT-5.4 mini gave Free and Go users a lighter reasoning option. GPT-5.4 is scheduled to retire from ChatGPT on July 23, 2026.
GPT-5.5 (April 23, 2026)
The current generally available flagship before GPT-5.6, and the biggest jump on agentic tasks since GPT-5. It handles multi tool workflows, writes and runs code, and performs web research across steps with less hand holding. The GPT-5.5 Pro variant targets the hardest science, finance, and legal work.
GPT-5.5 Instant (May 5, 2026)
The model most people talk to today. It replaced GPT-5.3 Instant as the default across every plan, cut hallucinated claims sharply on high stakes prompts in medicine, law, and finance, and can pull context from past chats, files, and connected accounts for more personal answers. Its free tier fallback is GPT-5.5 Instant Mini.
GPT-5.6 Sol, Terra, and Luna (July 9, 2026)
The newest release, covered in full above. Sol leads, Terra balances cost and quality, and Luna prioritizes speed and price, all under the new tier based naming system.
Comparison of current ChatGPT models
This table focuses on models that are active and relevant today, with OpenAI's published API rates per one million tokens. Older models such as GPT-4 and the original GPT-5 have been retired from ChatGPT, though some remain callable through the API for existing projects.
Model | Released | Role today | Input | Output |
|---|---|---|---|---|
GPT-5.6 Sol | Jul 2026 | Flagship reasoning | $5.00 | $30.00 |
GPT-5.6 Terra | Jul 2026 | Balanced everyday | $2.50 | $15.00 |
GPT-5.6 Luna | Jul 2026 | Fast and cheap | $1.00 | $6.00 |
GPT-5.5 | Apr 2026 | Prior flagship | $5.00 | $30.00 |
GPT-5.5 Instant | May 2026 | Default in ChatGPT | Varies | Varies |
GPT-5 | Aug 2025 | Legacy, API | $1.25 | $10.00 |
GPT-5 mini | Aug 2025 | Efficient | $0.25 | $2.00 |
GPT-5 nano | Aug 2025 | Lowest cost | $0.05 | $0.40 |
GPT-4o | 2024 | Legacy, API | $2.50 | $10.00 |
GPT-4o mini | 2024 | Legacy, API | $0.15 | $0.60 |
Prices can change, so confirm current rates on OpenAI's pricing page before you budget a project.
Why does ChatGPT have so many versions?
Two reasons, and both are practical.
First, OpenAI now ships smaller, more frequent updates instead of rare giant leaps. A point release like GPT-5.5 or GPT-5.6 improves specific areas, coding, reasoning, cost, without forcing a full rebuild of the product. That is why version numbers climb quickly.
Second, different jobs need different tradeoffs. A quick email does not need the same horsepower as a multi step security review, and running the flagship on every request would be slow and expensive. Splitting a generation into tiers like Sol, Terra, and Luna lets you match cost and speed to the task. The tradeoff is complexity, which is exactly why the picker got simplified into Instant, Thinking, and Pro for everyday users.
How honest are the benchmark numbers?
Worth a clear word here, because most roundups skip it. The eye catching scores you see, on Terminal-Bench, SWE-Bench, coding indexes, are almost always self reported by the model maker. They are a useful signal, not a verdict.
On OpenAI's own published tables for GPT-5.6, the picture is mixed rather than a clean sweep. Sol leads on several long horizon agentic tasks, the coding agent index, and computer use. At the same time, those same tables show Claude models leading on SWE-Bench Pro by a wide margin, where Sol's 64.6 percent trails Claude Mythos 5's 80.3 percent by roughly 15 points. If your real work is repository scale software engineering, that gap matters.
The reliable move is to pick the two or three tasks you actually care about, run your top candidate models side by side on your own inputs, and measure the outcome that counts for you, whether that is review time, error rate, or cost per finished task. Benchmarks narrow the field. Your own test picks the winner.
The bottom line
The short answer to "what is the latest version of ChatGPT" is GPT-5.6, the Sol, Terra, and Luna family that went live on July 9, 2026. The more useful answer is that GPT-5.5 Instant still handles your everyday chats, while GPT-5.6 Sol steps in for reasoning on paid plans. Once you understand that split, and the new pattern where the number is the generation and the name is the tier, the rest of the lineup stops feeling like alphabet soup.
If you are choosing a plan or comparing ChatGPT with other assistants, our related guides go deeper on the tradeoffs: see ChatGPT Free vs Paid, Perplexity vs ChatGPT, and Grok vs ChatGPT. You can also browse our full directory of AI chatbots and assistants to find the right tool for your workflow.