ChatGPT / GPT models: the default general-purpose LLM

OpenAI’s ChatGPT product and the GPT API family behind it. This is a model profile for chat, writing, coding, and agents — not a SKU changelog and not a Jev page.

What ChatGPT is

ChatGPT is OpenAI’s consumer assistant. Behind it sits a GPT model family you can also call from the API. Both generate tokens: answers, drafts, code, and tool calls. That is the job. If you needed a bounded label and nothing else, you are on the wrong profile — start with Jev or the classification shortlist.

This page is a family profile. OpenAI retires and renames SKUs. Treat “ChatGPT” as the product, and pin a dated API alias in production.

Developer / company

OpenAI. Consumer surface: ChatGPT. Builder surface: the OpenAI API, plus Azure-hosted copies. Biography is not a score.

Launch date and current versions

ChatGPT launched in November 2022. As of 20 September 2026, official API docs still list GPT-5 as a previous coding / reasoning model, GPT-5.6 Sol as a 5.6 flagship alias, and GPT-6 Astra as the recommended latest model. Those names are vendor aliases. They will move. Pin snapshots when a workflow is load-bearing.

Pricing (vendor claim)

Two different bills:

• ChatGPT consumer plans (Free / Plus / Pro and whatever bundle is current). Seat price, not token math.
• API tokens. You pay input and output. Example list rates published by OpenAI as of this desk date: GPT-5 $1.25 / $10 per 1M tokens; GPT-5.6 Sol $4 / $20. Cached input is cheaper. GPT-6 Astra is listed higher. openai.com/api/pricing is the source of truth — this paragraph is a dated snapshot, not a rate card.

For classify/route volume, compare the Jev pricing review. For “which chatbot should a person use,” price is usually a seat, not a token.

Input / output

• Input: text, plus images and other modalities depending on SKU and product surface.
• Output: generated text, code, and tool calls. Structured outputs / JSON schema are a constraint on generation, not a different model class.
• Reasoning-effort settings trade latency and cost for longer internal chains. Vendor claim; we have not published our own curve.

Main use cases

Chat, writing, coding, research help, multimodal Q&A, and tool-using agents. Classification, routing, and extraction are possible — you still generate tokens to get a label. That is fine at low QPS. It is the wrong default at high QPS. See Jev vs ChatGPT.

API availability and integrations

First-party API, Azure OpenAI, and almost every gateway catalog. Ecosystem score is the strongest row on this card. That does not make GPT the right model for a closed enum.

Strengths

• Default product people already have an account for.
• Mature tools, structured outputs, and third-party wiring.
• A full SKU ladder — you can usually buy a cheaper sibling before you leave the vendor.
• Public third-party evals exist, even if this desk has not rerun them.

Limitations

• Generation is the architecture. You cannot make it “not talk.”
• Input + output pricing makes high-QPS decisions expensive versus a decision model or a frozen classifier.
• Alias churn. “GPT-5” in a blog post from last quarter may already be a previous model.
• Most “#1 on leaderboard” screenshots are someone else’s benchmark, not yours.

When to pick something else

• Need a typed allow/deny or route and nothing else → Jev or a classical classifier.
• Need long-document caution as the default → Claude.
• Need Workspace / Search-native grounding → Gemini.

Latest updates

September 2026 desk note: GPT-6 Astra is the vendor-recommended flagship; GPT-5 remains listed as a previous model with a still-useful price point. This profile will date-stamp alias and price moves. It will not become a ChatGPT how-to.

Task fit, not a winner