Choosing a ChatGPT model used to mean sorting through names such as o1, o3, GPT-4o, and several smaller variations. In 2026, OpenAI replaced much of that clutter with a simpler set of options. It then introduced a second naming system for Work, Codex, and developers, because apparently complete simplicity was never on the table.
For most people, the answer is straightforward. Leave ChatGPT on Instant for everyday use. Move to Medium or High when a task requires more analysis, planning, research, or difficult coding. Extra High and Pro are for problems where you care more about getting the strongest answer than getting it quickly.
The confusing part is that these labels aren’t necessarily different models. According to OpenAI’s current ChatGPT model guide, Instant is powered by GPT-5.5 Instant, while Medium, High, and Extra High all use GPT-5.6 Sol with different amounts of reasoning. Pro uses GPT-5.6 Sol Pro.
Instant is the sensible default. You don’t need a frontier reasoning model to rewrite an email, suggest dinner recipes, or explain why your printer has once again chosen violence.
Medium is the best place to start when Instant produces a shallow answer or when the task has several moving parts. High is useful when the model needs to inspect more evidence, resolve conflicting requirements, or work through a technical problem carefully.
Medium, High, and Extra High aren’t separate models. All three use GPT-5.6 Sol, with the setting determining how much reasoning effort it applies. Pro is different because it uses GPT-5.6 Sol Pro, which OpenAI positions as its highest-capability choice for difficult tasks and longer workflows.
Eligible paid accounts can also let ChatGPT decide when to move from Instant to Medium. Open the model picker and select Configure to turn automatic switching on or off. Leaving it enabled is the easiest option for most users because ChatGPT can stay fast for simple requests and apply more reasoning when a prompt appears to require it.
Automatic switching won’t necessarily give every complicated-looking prompt the maximum possible reasoning. If the first result still feels superficial, manually selecting High can help. Extra High and Pro should be reserved for genuinely difficult work. Selecting them for every prompt may simply make an easy answer arrive later.
Free accounts can do considerably more than basic text chat. They include search, voice, projects, plugins, and access to public GPTs, along with limited uploads, data analysis, image generation, deep research, Codex, and ChatGPT Work. The biggest differences between plans are the available models, usage limits, and whether the account includes individual or organizational tools.
OpenAI’s Free plan is capable enough for occasional questions, research, uploads, and image generation. Its separate usage limits are the catch. You may still be able to chat after reaching an image, upload, or data-analysis limit, but you’ll have to wait before using that particular tool again.
ChatGPT Go is aimed at people who regularly hit those Free limits but don’t need the strongest models. It provides more messages, images, uploads, data analysis, and memory. It can also include projects, tasks, custom GPTs, and Library access, although feature availability and storage limits still apply.
Go doesn’t unlock GPT-5.6 Sol in a normal ChatGPT conversation. It does, however, include limited GPT-5.6 Terra access through Work and Codex on desktop. That makes it more than Free with a slightly larger message bucket, but it remains below the advanced reasoning options included with Plus.
ChatGPT Plus is the important dividing line for individual users. It unlocks GPT-5.6 Sol at Medium and High, along with expanded access to deep research, Codex, Work, scheduled tasks, connected apps, and custom GPT creation. For most people using ChatGPT regularly for professional work, Plus is the most sensible paid tier.
Pro is aimed at people who use those tools heavily or need Extra High and GPT-5.6 Sol Pro. OpenAI offers two Pro tiers. The $100 subscription provides five times the usage allowance of Plus, while the $200 subscription provides 20 times the allowance. Their core capabilities are otherwise the same.
Business and Enterprise aren’t simply more powerful individual subscriptions. ChatGPT Business adds a managed workspace, company knowledge, connected internal tools, administration, spend controls, and business-data protections. ChatGPT Enterprise adds more advanced identity, governance, compliance, retention, data-residency, and support controls for larger deployments.
Exact allowances can vary with demand, workspace settings, and OpenAI’s current policies. The ChatGPT pricing page and the model picker on your account provide the most current picture of what you can actually use.
Sol, Terra, and Luna appear in ChatGPT Work, Codex, and the OpenAI API. They aren’t three more choices hidden inside a regular ChatGPT conversation.
The first decision is which experience matches the job.
ChatGPT Work can work across files, connected apps, and tools over a longer task. Codex applies a similar agentic approach specifically to software development. Both can plan and perform several steps instead of returning only a single chat response.
Codex isn’t powered by one fixed model. Its recommended lineup is the GPT-5.6 family.
GPT-5.6 Sol is the flagship. It’s intended for difficult, open-ended work that requires deeper analysis, judgment, or polish, including complex code changes, research, design work, and technical analysis.
GPT-5.6 Terra is the practical all-rounder. It balances capability, speed, and cost for everyday coding, document analysis, reporting, and other production work.
GPT-5.6 Luna is the fastest and most economical option. It makes sense for clear, repeatable jobs such as extraction, classification, transformation, focused coding, and high-volume tasks.
The default Power setting in Codex uses GPT-5.6 Sol at Medium reasoning. You can move toward higher reasoning for more difficult tasks or choose a faster model when the work is narrow and predictable. OpenAI’s Codex model guide recommends starting with Sol when you’re unsure.
Free and Go accounts receive limited access to Terra in Codex. Plus, Pro, Business, and Enterprise can use Sol, Terra, and Luna, subject to plan and workspace limits. Pro also includes GPT-5.3-Codex-Spark, a text-only research-preview model designed for near-instant coding iteration.
Codex can also be used with an API key. In that case, model access and billing follow the associated API project rather than the user’s ChatGPT subscription. API-key access also doesn’t include subscription features such as cloud code review or Slack integration. OpenAI maintains separate documentation for Codex plans and pricing.
The ChatGPT model picker and the OpenAI API model catalog are separate. Paying for ChatGPT Plus or Pro doesn’t include API usage, and a model disappearing from ChatGPT doesn’t necessarily mean it has been removed from the API.
For general-purpose API projects, the current GPT-5.6 choices are:
Sol is the starting point when capability is the priority. Terra will be more practical for many production applications because it lowers operating costs without dropping to the smallest model tier. Luna is designed for workloads where requests are frequent, relatively predictable, and highly sensitive to price or latency.
Developers may also need specialized models for image generation, speech, transcription, embeddings, moderation, or real-time conversations. Those models solve different problems and shouldn’t be treated as direct alternatives to Sol, Terra, or Luna.
API prices and limits change regularly, so hard-coding them into a general model guide would give this page the shelf life of an open carton of milk. Check OpenAI’s current model pages before choosing a production model.
If you arrived comparing o1 with o3-mini, neither is the sensible starting point for a new ChatGPT workflow in 2026. The original version of this guide focused on models that have since disappeared from the normal ChatGPT picker.
GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini were retired from ChatGPT in February 2026. OpenAI o3 remains available to some paid users through legacy model settings, but it is scheduled to leave ChatGPT on August 26, 2026.
Older conversations generally continue using a newer corresponding model after a retirement. You don’t need to recreate every chat simply because its original model has disappeared.
These ChatGPT retirements also don’t automatically apply to developers. OpenAI has kept some older models available through the API after removing them from ChatGPT, with separate notices issued before an API model is deprecated. Anyone maintaining an application should check the API catalog and deprecation notices instead of assuming that changes to the consumer model picker apply to their code.
The older models remain important to ChatGPT’s history. GPT-4o brought faster multimodal interaction, while o3 helped establish the reasoning controls that eventually evolved into today’s model picker. They just aren’t where most people should begin now.
Stick with Instant for casual use. Start with Medium for research, professional writing, planning, data analysis, or moderately difficult code, then move to High when the task includes dense evidence, conflicting requirements, or a stubborn technical problem. Extra High and Pro are best reserved for work that genuinely benefits from more reasoning.
In Work or Codex, Terra is the practical everyday choice, Sol is for the hardest jobs, and Luna is for speed and efficiency. You can also leave automatic switching enabled and let ChatGPT handle most routine model decisions. The picker is useful, but it doesn’t need to become another productivity system you spend more time managing than using.
If OpenAI’s lineup still feels needlessly complicated, it isn’t the only option. Our guide to the best AI chatbots compares ChatGPT with its biggest alternatives.
GPT-5.6 Sol is limited to eligible paid plans and remains subject to rollout and workspace settings. Confirm that you’re signed in to the correct account and check whether an administrator controls model access. Free, Go, and logged-out users don’t receive Sol in standard ChatGPT conversations.
No. Instant is designed for everyday questions, and eligible paid accounts can allow ChatGPT to switch automatically to Medium when a request needs more reasoning. Manual selection becomes useful when a task is unusually complex or the first answer isn’t thorough enough.
No. All three use GPT-5.6 Sol. The setting changes how much reasoning effort the model applies. Pro uses the separate GPT-5.6 Sol Pro model.
Start with the default Power setting, which uses GPT-5.6 Sol at Medium reasoning. Terra is better for everyday work where speed and efficiency matter, while Luna suits narrow, repeated, or high-volume tasks.
Their removal from ChatGPT doesn’t automatically remove them from the API. GPT-4o remained available through the API after its ChatGPT retirement, and OpenAI has said o3’s August 2026 retirement applies to ChatGPT only. Developers should check the current API catalog and deprecation notices before migrating.
ASUS has started selling its UGen Intel Arc Pro B70 graphics card in Europe, bringing its take on Intel’s workstation GPU to customers with a hefty €1,799 recommended price. The card is available in Germany and Austria, while ASUS lists it at CHF 1,799 in Switzerland.
Intel first introduced the Arc Pro B70 in March 2026, with its own branded model starting at $949. ASUS’s version arrives several months later with the same core GPU architecture but a different cooling design and a higher power rating. The price also places it well above Intel’s original starting point, although the two prices are not directly comparable because ASUS’s European figure includes VAT.
Anthropic has officially filed to go public, and buried inside its S-1 is a blunt warning about the technology at the center of its business. The maker of Claude AI says advanced AI could cause everything from large-scale cyberattacks and geopolitical instability to catastrophic or even existential risks for humanity.
The warnings take up a surprisingly large part of the filing. Axios notes that more than 90 pages, or roughly a third of the prospectus, focus on risk factors. Many of them warn about what could happen as AI systems become more capable and autonomous.
AI giants have a growing autonomy problem on their hands. A new Axios report has revealed that Anthropic's Claude recently accessed publicly available Georgia vote count without human authorization, sparking fresh concerns about AI security ahead of the upcoming midterm elections.