ChatGPT Pricing in July 2026:
Free, Go, Plus, Pro, Business,
Enterprise and API Costs

Official ChatGPT pricing for Go, Plus, Pro and Business plans in July 2026

ChatGPT costs from $0 to $200 per month for individuals, plus per-seat business plans. Free is $0 with ads. Go is $8/month, Plus is $20/month, and the two Pro tiers are $100 and $200/month. ChatGPT Business is $20 per seat on annual billing ($25 monthly), and Enterprise is custom. The flagship API model, GPT-5.5, is $5 per million input tokens and $30 per million output.

Full breakdown of each tier, the model-routing gap, and the complete API rate table below.

This is the question every ChatGPT user has asked at least once, and every comparison article skips. The model picker shows three labels – Instant, Thinking, Pro – and the underlying model is selected automatically. To see which model handled a query, you have to open a Configure setting most people never touch. API users always get the specific model ID in response metadata. ChatGPT users on default settings do not.

Why this matters for what you pay: a Plus subscriber pays $20 a month expecting GPT-5.5, because that is the headline model. The Auto selector may route a query to GPT-5.5, GPT-5.4, or a smaller fast variant based on logic OpenAI has not fully documented. On a hard prompt, you get the flagship. On a routine one, you may not. The price is fixed. The value per query is not.

Since February 9, 2026, the US Free tier shows advertisements – the first time OpenAI has placed ads in ChatGPT. The tier runs GPT-5.3 Instant, capped at roughly 10 messages per 5-hour window before it falls back to GPT-5.3 Mini. The non-reasoning context window is 16K tokens, a fraction of Plus, which is the single clearest reason to upgrade for document work.

The real cost of Free shows up the moment you act on an answer. Per the Suprmind Multi-Model Divergence Index (April 2026 Edition, n=1,324 production turns), ChatGPT’s catch ratio is 0.38 – it gets caught by other models 295 times for every 111 corrections it makes. On Free without web search enabled, the citation hallucination rate reached 67% in the Columbia Journalism Review audit cited in Suprmind’s AI Hallucination Rates and Benchmarks reference. Free is fine for casual queries. It is not fine for citation-dependent work, even at a price of zero.

Go launched globally on January 16, 2026 after an August 2025 India-only debut. It runs GPT-5.2 Instant with GPT-5.3 access at a message cap, gives roughly 10x Free limits across messages, uploads, and image creation, and adds expanded memory. Codex Mobile is included as a free preview. The catch: Go still shows ads. Paying $8 a month does not buy out the advertisement layer.

Plus is the entry point for serious use. It includes GPT-5.5 Instant and GPT-5.5 Thinking through the Auto selector, GPT-5.4 Pro and o3 in Flexible mode, 80 file uploads per 3-hour window, 25 files per Project, 10 Deep Research queries per month, Advanced Voice Mode, image generation, ChatGPT Agent mode, Canvas, Tasks, Sora, and Custom GPT creation. No ads. The price has held at $20 for three years.

Plus is monthly-only. There is no annual discount, and the $48-per-year figure that circulated earlier in 2026 was never a standard published rate. If you see it quoted, treat it as a lapsed or region-specific promotion, not a plan you can sign up for today.

The constraining number for research-heavy work is the 10 Deep Research queries per month. Each session runs 5 to 30 minutes and reads dozens of pages. Ten a month is two or three serious deliverables at most. If your work demands more, the Pro tiers are the answer.

See also: ChatGPT Deep Research deep dive →

OpenAI added a $100 Pro tier on April 9, 2026, sitting between Plus and the existing $200 Pro. Both include GPT-5.5 Pro and the same core feature set, plus the US Personal Finance preview that links 12,000+ banks through Plaid. The differences are quantitative.

ChatGPT Business (formerly ChatGPT Team, renamed August 2025) dropped to $20 per seat on annual billing, or $25 monthly, effective April 2, 2026 – a $5 cut on both from the old $25 and $30. The minimum is two seats.
Business runs GPT-5.5, and adds shared workspaces, SAML SSO, admin controls, SOC 2 Type 2, ISO/IEC 27001, 27017, 27018, and 27701, no model training on your data, the Codex agent, Deep Research, and 60+ connectors. SCIM provisioning stays Enterprise-only. Nonprofits can get a substantial discount through OpenAI for Nonprofits.

At two or more seats, ChatGPT Business often costs the same per person as Plus while adding governance and compliance, which is why the crossover point is lower than most teams assume. The constraining limit is context on non-reasoning models. For teams routinely working with very large documents or codebases, the window becomes the bottleneck before the per-seat price does.

OpenAI Enterprise Plan

Enterprise is custom-priced, generally aimed at organizations of roughly 150+ users, with independent estimates in the $60 to $100+ per seat range. It adds SCIM provisioning, enterprise key management, role-based access control, an analytics dashboard, IP allowlisting, data residency across the US, EU, UK, JP, CA, KR, SG, IN, AU, and UAE, a global admin console, 24/7 priority support, and custom legal terms.

For procurement teams: Enterprise is the only tier where you can negotiate around model-routing transparency, data retention, and custom SLAs. If you carry a regulated compliance posture or a data residency requirement, the negotiated terms matter more than the per-seat math.

The API runs the same GPT models that power ChatGPT, but bills per token instead of per seat. For workloads with predictable cost-per-query, it can be far cheaper than a Plus seat. For high per-query token volume, it can be far more expensive. Pricing is per million tokens.

Source: platform.openai.com/docs/pricing, verified July 8, 2026. The “272K” figure sometimes quoted for GPT-5.4 is a pricing inflection point, not a context cap – the model runs to 1M tokens. o4-mini was retired from the ChatGPT interface in February 2026 but remains available on the API. Rates change often, so confirm at the docs before budgeting.

The API also offers three modifier tiers. Batch processing runs at 50% of standard rates with 24-hour async turnaround. Flex trades lower cost for slower responses and occasional resource unavailability. Priority is pay-as-you-go at 2.5x standard rates for guaranteed throughput.

GPT-4.1 nano at $0.10 per million input tokens is 50x cheaper than GPT-5.5 at $5.00. GPT-4o mini at $0.15 lands at 33x. On output the gap is wider still. For workloads that do not need flagship capability, the small models are the cost-efficient default, and the ratio is larger now than it was three months ago because the nano tier keeps getting cheaper.

When does the cheap model win? High-volume routine work: classification, extraction from structured data, content moderation, simple summarization, anything you can verify with rules rather than trust. The hallucination profile matters less when the task is constrained. GPT-4o mini’s Vectara old-dataset hallucination rate of 1.7% is lower than GPT-5’s rate above 10% on the harder Vectara new dataset. For narrow workloads, the smaller cheaper model is often the safer choice too.

The case for paying 50x more: open-domain reasoning, multi-step planning, agentic computer use, long-context recall, and any work where the question itself changes shape during the conversation. Flagship capability earns the price. Routine work does not.

Cached inputs cost 90% less than fresh inputs across the GPT-5.x family. GPT-5.5 cached input is $0.50 per million versus $5.00 fresh. GPT-5.4 cached is $0.25 versus $2.50. Caching applies when a request reuses a substantial portion of a recent prior request’s prompt, and the API caches frequently used prefixes automatically.

Workflows that benefit most: agentic loops with consistent system prompts, RAG pipelines reusing the same document context, support bots with stable persona instructions, any production workload with templated prompts. If your spend is dominated by input tokens and your prompts carry stable prefixes, caching is the single largest lever for cost control, with no engineering work beyond structuring the prompt.

Fourteen material pricing or model changes in the twelve months to July 2026, six of them in the last 90 days. Any comparison built today needs re-verification before a decision over $1,000 a month. That volatility is exactly why a single-model view is fragile: the model behind your seat can change without notice, and so can what it costs.

ChatGPT is available in most countries. Sanctioned jurisdictions (Iran, North Korea, Cuba, and in some cases Russia) are blocked under US export-control compliance. Mainland China is unavailable. The UK, EU, and most of Asia and Latin America have access.

The Italian data protection authority temporarily banned ChatGPT in March 2023 over GDPR concerns. OpenAI complied within the deadline by adding privacy disclosures, age verification, and a training opt-out tool, and service was restored in May 2023 without a formal fine. The episode established that EU data protection authorities can act against AI systems without waiting for EU AI Act enforcement, which is a procurement risk to factor into European deployments at scale.

For Enterprise customers, OpenAI offers data residency in the US, EU, UK, JP, CA, KR, SG, IN, AU, and UAE. If you have a regulatory obligation to keep data in a specific jurisdiction, Enterprise is the tier where that becomes negotiable.

FAQ

ChatGPT Pricing: Frequently Asked Questions

Sources

• openai.com/chatgpt/pricing and chatgpt.com/pricing (consumer pricing)
• platform.openai.com/docs/pricing (API pricing)
• openai.com/index (model launch and product announcements)
• Suprmind Multi-Model Divergence Index (catch ratio data)
• Suprmind AI Hallucination Rates and Benchmarks (per-model hallucination data)

Last verified July 8, 2026. ChatGPT pricing changes more often than any other AI product – confirm at openai.com/chatgpt/pricing before making decisions.