ChatGPT versions explained: models, plans and API costs
Updated October 22, 2026 · first published October 22, 2026
“ChatGPT versions” can mean three different things: your subscription plan, the model answering, or the product surface you are using. Free, Plus and Pro are plans. GPT-6 and GPT-6.1 Sol are model names. Chat, Work, Codex and the API have different access and billing rules. Check those separately before upgrading or estimating costs.
This guide explains those distinctions and gives a worked budget for a team using both subscriptions and API calls. Product details and USD price references were checked against official OpenAI documentation on 22 October 2026. Availability can vary by account, region, rollout and workspace policy.
What does each ChatGPT name actually identify?
| Name or control | What it identifies | What it does not establish |
|---|---|---|
| Free, Go, Plus, Pro, Business | A subscription or workspace plan | The model behind every request |
| GPT-6, GPT-6.1 Sol, GPT-6 Astra | A model or model family | Your subscription price or unlimited access |
| Instant, Medium, High, Extra High | A response or thinking setting in the relevant interface | A separate paid subscription |
| Chat, Work, Codex | The product experience handling the task | Identical model access and usage rules |
| API model ID and service tier | The model and processing mode selected by software | Usage included in a ChatGPT seat |
The most expensive misunderstanding is treating a plan name as a model entitlement everywhere. A Pro subscription and the Pro thinking option are different concepts. Likewise, a model visible in Work does not prove it is selectable in Chat or accessible to your API organization.
Which ChatGPT plans are free or paid?
OpenAI's published pricing guide lists these USD reference prices. Treat the displayed checkout, billing period and applicable taxes as the purchase terms for your account.
| Plan | Published reference | Budgeting question |
|---|---|---|
| Free | $0 per month | Do the available tools and limits cover the task? |
| Go | $8 per month | Does this account need a higher allowance or different capabilities? |
| Plus | $20 per month | Is the included access sufficient for this user's work? |
| Pro | $100, $200 or $500 per month | Which tier's actual limits and capabilities justify the extra seat cost? |
| Business | $20 per user per month billed annually; $25 billed monthly, with 2+ users | Are shared workspace controls and seat commitments required? |
| Enterprise or Edu | Contact sales | What does the agreement include, and how are additional credits billed? |
The plan price is not a promise of unlimited use of every model, tool or speed setting. For purchasing, record the seat price, billing term, included allowance, additional-credit policy and named owner. A team can spend more through unused seats and unnecessary upgrades even while its API unit prices fall.
Which models are in Chat versus Work and Codex?
OpenAI's Chat model guidance describes GPT-6 rolling out to eligible paid Chat accounts. When GPT-6 is selected, Sol powers Instant through Extra High, while the Pro option uses GPT-6 Pro, powered by Astra. Which controls appear depends on the plan and workspace. Plus access to Astra in Work or Codex does not mean Plus includes the Pro option in Chat.
For Work and Codex, the official model guide separately documents GPT-6.1 Sol and GPT-6 Luna availability. It describes GPT-6.1 Sol access as dependent on plan, client and rollout, with administrator controls in managed workspaces. Do not infer a universal model list from another person's screenshot.
Free and Go accounts can show Luna-family defaults during rollouts. Older explicitly selected models can also remain available where supported. Use the model picker in the exact product and account you intend to use; the release notes explain rollout changes, rather than guaranteeing your account has already received them.
How can you check which version you are using?
- Identify the account and workspace: a personal subscription and an employer-managed workspace may have different permissions.
- Identify the surface: distinguish an ordinary Chat conversation from Work, Codex or your own API-backed application.
- Read the active model and thinking controls: a saved explicit selection can differ from the current default.
- Check the allowance separately: use the account's usage and billing views, including any additional credits.
- For software, log the returned model: keep the requested model, returned model and service tier with the usage record.
Asking the assistant to identify itself is not a billing record. For a team inventory, maintain one row per product and account owner: plan, active model policy, monthly commitment, usage limit, renewal date and billing owner.
Does ChatGPT Plus or Pro include API credits?
OpenAI documents separate ChatGPT and API billing systems. Paid API usage is billed separately from your ChatGPT subscription. A developer application using an API key needs its own billing setup and model access.
Keep three ledger categories: recurring seats, additional ChatGPT credits, and metered API consumption. They are not interchangeable balances. The Work/Codex pricing guide also describes shared usage between those surfaces; moving a task between them does not necessarily create a fresh allowance.
What do current API models cost per task?
The following short-context standard rates come from OpenAI API pricing, in USD per million tokens. They are examples for metered API budgeting, not a conversion table for subscription messages.
| API model | Ordinary input | Cached input read | Output |
|---|---|---|---|
| GPT-6 Luna | $0.10 | $0.01 | $0.50 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6 Astra | $10.00 | $1.00 | $50.00 |
Illustrative workload: 10,000 single-call requests per month, each with 4,000 ordinary input tokens and 1,000 billed output tokens. That is 40 million input and 10 million output tokens.
| Model | Monthly calculation | Modeled token cost |
|---|---|---|
| Luna | 40 × $0.10 + 10 × $0.50 | $9 |
| 6.1 Sol | 40 × $2 + 10 × $10 | $180 |
| Astra | 40 × $10 + 10 × $50 | $900 |
These calculations hold token counts constant to isolate the rates. They do not establish equal quality or real completion cost. Models can use different amounts of reasoning, require different retries, or fail different tasks. Include all billable output and compare cost per accepted result on your own workload.
Cache writes, long-context bands, different processing tiers, paid tools, infrastructure and taxes are excluded. Their rates must be applied where relevant. Our Astra API cost guide explains disjoint cache categories and context thresholds.
How should you budget seats and API usage together?
Suppose ten users each need a $20 monthly Plus seat, while a customer-facing feature runs the illustrative Sol workload above. The budget is $200 for seats plus $180 for API tokens: $380 per month before credits, tools and other costs. Purchasing the seats does not remove the API bill.
Allocate seats to their human owners and API requests to the team, feature and tenant causing them. Review inactive seats and API spend separately. When choosing a more expensive model, test whether improved success or reduced review time justifies the complete cost rather than upgrading every user and request together.
Why do older model names still appear?
A product-specific retirement is not necessarily a global API retirement. OpenAI's model guide, for example, distinguishes GPT-5.5 retirement from Codex with ChatGPT sign-in from API-key availability. Read the exact product and authentication scope in a notice before changing a production integration.
For operational planning, keep the model identifier and billing source in configuration. Set a review owner for deprecations and re-run a small acceptance set before migrating. See budgeting for forced model migrations for that process.
Which option should you choose first?
For personal interactive use, start with the lowest plan that provides the tools and allowance you need. For a managed team, include administration and data requirements in the seat decision. For software serving customers, evaluate API models against a fixed acceptance test and an explicit cost budget.
The useful purchasing unit is the job completed at the required quality. Keep the subscription decision separate from the API routing decision, and connect both to a named budget owner.
Related
- GPT Astra API pricing and cache costs
- Estimate monthly LLM spend
- Budgeting for model migrations
- Understanding SaaS AI credits
Want this applied to your stack? Bring the provider bills, gateway logs, and top workflows; we will map the cost drivers and savings path. Book a free audit →