OpenAI pricing

About

OpenAI is a San Francisco-based AI research and deployment company, founded in 2015 by Sam Altman, Greg Brockman, Ilya Sutskever, Elon Musk, and others. Originally a nonprofit, it transitioned to a “capped-profit” structure in 2019 and completed a full for-profit conversion in 2025. OpenAI is widely considered the most consequential AI company of the current era: it launched ChatGPT in November 2022, kickstarting the consumer AI boom, and has consistently pushed the frontier of large language model capability and deployment at scale.

By mid-2025 OpenAI reports over $5B in annualized revenue, over 300 million weekly active ChatGPT users, and a valuation exceeding $300B following a secondary share sale. The company has raised over $20B from investors including Microsoft ($13B committed), Tiger Global, Sequoia, Andreessen Horowitz, and Thrive Capital. Microsoft’s partnership gives OpenAI exclusive access to Azure compute at preferential rates and integrates GPT models across Microsoft 365, Azure AI, and GitHub Copilot.

OpenAI competes with Anthropic, Google DeepMind, Meta AI, Mistral, and increasingly with DeepSeek on both the consumer (ChatGPT vs. Claude.ai, Gemini, Copilot) and API fronts. Its pricing history is the defining cost-deflation story of the AI era: from $60/1M output tokens at GPT-4 launch in March 2023 to equivalent capability for under $1/1M by mid-2025.

Pricing summary : How OpenAI’s consumer subscription and API pricing work together

OpenAI runs two parallel pricing surfaces that serve different customer profiles with minimal overlap. The consumer subscription stack (ChatGPT Free → Go → Plus → Pro → Business → Enterprise) is flat-rate monthly/annual, gated by model access quality and usage quotas. The API is pure pay-per-token with no subscription minimum — billed to a credit-card-charged prepaid balance.

Consumer pricing follows a freemium expansion model where the free tier drives mass adoption, the $8 Go tier provides a low entry price point, and the $20 Plus tier captures the majority of individual revenue. The Pro tier now spans $100/month and $200/month, targeting power users who need progressively higher limits and Pro-mode compute rather than serving as a linear upsell from Plus.

The API’s cost curve has fallen steeply since 2023, driven by model efficiency improvements, hardware cost reductions, and competitive pressure from open-source and Chinese rivals. This deflationary dynamic is core to OpenAI’s strategy: lower prices expand the API market faster than any sales motion could.

What makes this different: The GPT-5.4 nano model at $0.20/1M input tokens continues the commoditization of intelligence — a fraction of the price of GPT-4 at launch. This changes the economics for any business building on AI, shifting the constraint from model cost to product and distribution.

Pricing by product

ChatGPT Consumer Subscriptions

Business is $20/seat/mo on annual billing; $25/seat/mo on monthly billing. Pro is offered at both $100/mo and $200/mo.

API — GPT-5.x Text Models (Current Generation)

Pricing reflects standard processing for context lengths under 270K. Batch API: 50% off input and output. Prompt caching is billed at the reduced cached-input rate shown above.

API — Tools & Service Tiers

Sales motions across products: PLG / self-serve for Free, Go, Plus, Pro, Business, and all API plans; sales-led for Enterprise. Prices accessed 2026-05-30.

Hidden costs : What OpenAI users actually pay beyond the base plan

Archetype A: Developer building a customer-facing chatbot on GPT-5.4

A developer with a 10,000 user chatbot averaging 500 input tokens + 300 output tokens per session, running 3 sessions/user/day:

This cost range illustrates why model selection and caching strategy are the primary cost levers for AI product companies. The AI cost unpredictability problem is real: a 10K-user product can easily generate six-figure monthly API bills without architectural discipline.

Archetype B: 50-person team on ChatGPT Business (annual billing)

Business billing is transparent, but the lower per-seat rate only applies on annual commitment — monthly billing at $25/seat adds overhead. There is no mid-term seat reduction on annual contracts.

Use the OpenAI pricing calculator to estimate your monthly API cost based on model, token volume, and caching strategy.

Pricing evolution : OpenAI’s pricing history from GPT-3 beta to the GPT-5.x generation

Cadence

Tracked range: 2020 Q2–2026 Q2. Quarters not listed above were verified stable.

Notable changes

• 2023-02-01 — ChatGPT Plus launched at $20/month; first consumer revenue for OpenAI.
• 2023-03-14 — GPT-4 API launched at $60/1M output tokens — the highest-priced consumer AI model to date; created a premium API market segment.
• 2023-11-06 — GPT-4 Turbo announced at DevDay: $10/1M input (vs $30 for GPT-4), 128K context. Kickstarted the API cost-deflation cycle. (OpenAI DevDay announcement)
• 2024-05-13 — GPT-4o launched at $5/$15 per 1M (later reduced to $2.50/$10); first multimodal frontier model with competitive API pricing.
• 2024-07-18 — GPT-4o mini at $0.15/$0.60 per 1M tokens: 97× cheaper than GPT-4 at launch, broader performance than GPT-3.5 Turbo. Commoditized the “good enough for production” tier.
• 2024-12-05 — ChatGPT Pro launched at $200/month with unlimited o1 Pro mode. (OpenAI blog)
• 2025-04-14 — GPT-4.1 family launched with 1M context windows and sharpened coding performance; GPT-4.1 nano at $0.10/1M became the cheapest capable OpenAI model at the time. (OpenAI announcement)
• 2026 (current) — The GPT-5.x generation is the live API lineup: GPT-5.5 ($5/$30 per 1M), GPT-5.4 ($2.50/$15), GPT-5.4 mini ($0.75/$4.50), GPT-5.4 nano ($0.20/$1.25). On the consumer side, ChatGPT added a Go tier ($8/mo) and a second Pro tier ($100/mo alongside $200/mo), and the team plan was renamed Business ($20/seat annual).

What’s unique : OpenAI’s distinctive pricing mechanics

1. The fastest API cost deflation in technology history. GPT-4 launched at $60/1M output tokens in March 2023. By the GPT-5.x generation in 2026, equivalent-or-better capability is available far cheaper — GPT-5.4 nano outputs at $1.25/1M, and even flagship GPT-5.5 output ($30/1M) is half the original GPT-4 rate. No other technology stack — cloud compute, storage, bandwidth — has deflated this fast. This makes AI infrastructure cost modeling extremely difficult: a budget set in 2023 for GPT-4 calls may be wildly overstated against current models.

2. Dual-surface pricing that avoids consumer/developer cannibalization. The ChatGPT subscription and OpenAI API are priced entirely independently. A developer calling GPT-5.5 at $30/1M output tokens pays a token-metered rate with no relation to a flat ChatGPT seat — a dramatically different cost structure that prevents direct arbitrage. This dual-surface separation lets OpenAI optimize each surface independently for willingness-to-pay without one tier subsidizing the other.

3. A flagship-to-nano efficiency ladder within one generation. The GPT-5.x line spans GPT-5.5 ($5/$30) down to GPT-5.4 nano ($0.20/$1.25) — a roughly 25× spread in input price across one model family. This lets OpenAI serve frontier coding work and high-volume classification from the same API surface, with reduced cached-input rates and a 50% Batch discount layered on top. This matrix pricing model — capability tier cross-multiplied with efficiency tier — is more complex than any prior API pricing regime.

4. Batch API as a structural cost lever. The Batch API’s 50% discount for async workloads creates a two-class API market: latency-sensitive (full price) vs. batch/background (50% off). For developers doing classification, embedding, or bulk analysis, the Batch API changes unit economics dramatically. This time-of-use pricing approach is borrowed from cloud infrastructure and is increasingly standard in AI APIs.

5. Prompt caching as an invisible infrastructure subsidy. Cached input tokens cost 50% less. For applications with large, repeated system prompts (e.g., RAG contexts, tool manifests), this can reduce effective input costs by 30–60%. Because the cache is invisible to users — it’s an automatic benefit when the same prefix is reused — it rewards application architectures that maintain consistent system prompt prefixes.

Strengths & weaknesses

Billing UX : OpenAI’s subscription controls and API payment experience

• Consumer subscriptions — Managed in ChatGPT account billing settings. Plans can be upgraded/downgraded in-app; no sales call required for Go, Plus, Pro, or Business.
• API prepaid credits — API usage is billed against a prepaid balance charged to credit card or via invoice (Enterprise). Users can set a monthly spending limit to cap charges.
• Soft spend limits — Default is no hard limit; users must manually configure a monthly budget in the API billing settings, after which OpenAI stops serving requests. This is a frequent source of surprise bills from agentic workflows.
• Annual vs monthly toggle — Business is offered on annual or monthly billing (annual is the lower per-seat rate); Go, Plus, and Pro are monthly.
• Auto-renewal — All subscriptions auto-renew. Cancellation takes effect at end of the current billing period.
• Usage dashboard — API dashboard shows per-model token consumption, cost breakdown, and rate limit usage in near-real-time.
• Batch API invoicing — Batch jobs are invoiced on completion of the batch, not on a per-request basis. Large batches may take up to 24 hours to complete.
• Enterprise billing — Invoice-based, with custom payment terms. Microsoft Azure customers can access OpenAI via Azure AI Services under their Azure commitment.
• Free trial credits — New API accounts have historically received a small amount of free trial credits; the exact amount has varied by period and is not a published current figure.

Strategic wins : Why OpenAI’s pricing decisions worked

1. Free ChatGPT created a consumer distribution moat no API-first company can replicate

By launching ChatGPT as a free consumer product in November 2022, OpenAI built a 300M-user direct distribution channel before any competitor. This PLG flywheel means OpenAI reaches individual users, developers, and enterprise decision-makers through a single product surface. No API-first competitor (Anthropic, Mistral, Cohere) has matched this consumer footprint, creating an acquisition cost advantage that compounds with every viral ChatGPT moment. The free tier is not a cost center — it is the world’s largest AI marketing channel.

2. The $20 Plus anchor priced AI subscriptions for the long term

ChatGPT Plus at $20/month, launched February 2023, established the de facto price anchor for consumer AI subscriptions. Every subsequent competitor — Perplexity Pro, Claude Pro, Gemini Advanced — launched at $20/month or close to it. By being first, OpenAI defined the market’s perceived fair value for a premium AI subscription. This anchor pricing effect means even when competitors match features, they match OpenAI’s price rather than undercutting it.

3. Continuous API price cuts expanded the market faster than any sales motion

Rather than extracting maximum margin on GPT-4 (and protecting it from competition), OpenAI repeatedly cut API prices to expand the addressable market. Each price cut (GPT-4 Turbo, GPT-4o, GPT-4o mini) unlocked a new tier of developers who couldn’t previously afford production AI. The GPT-4o mini launch at $0.15/1M input created a production AI tier accessible to any SaaS company regardless of scale. This usage-based pricing expansion strategy mirrors AWS’s cloud pricing playbook and has been commercially validated by OpenAI’s ARR growth from ~$1B in 2023 to $5B+ in 2025.

4. The reasoning model premium tier created a second revenue dimension

By launching o1, o3, and o4-mini at significant premiums (o1 is 30× more expensive than GPT-4o per output token), OpenAI added a capability-based premium tier that its consumer subscription can’t fully capture. Enterprise developers building complex reasoning pipelines pay substantially more per token while still being on the same “usage-based” billing model. This tiered model pricing lets OpenAI simultaneously compete on low-cost commodity use cases (GPT-4.1 nano) and premium reasoning tasks (o1) with the same billing infrastructure.

Areas to improve : Gaps in OpenAI’s pricing approach

1. Default soft billing limits create runaway cost risk for agentic workloads

OpenAI’s API has no default spending cap. A developer deploying an agentic loop with a GPT-5.x model that runs unbounded can accumulate thousands of dollars in charges before receiving an alert. AWS, GCP, and Azure all provide configurable budget alerts and hard caps by default. OpenAI requires manual configuration in the dashboard. As agentic AI workflows become the norm, this cost unpredictability risk is the single most common cause of negative developer experiences. The fix is simple: default monthly spend caps, with easy override, and proactive email alerts at 50%/80%/100% of threshold.

2. Enterprise pricing opacity creates evaluation friction for procurement

ChatGPT Enterprise has zero published pricing. Procurement teams at enterprise companies cannot self-qualify or budget without a sales conversation. This sales-only price model creates unnecessary friction for budget-conscious buyers who need a rough number before engaging sales. Even a published range (“Enterprise starts at $X/seat/year for 50+ seats”) would accelerate evaluation cycles without sacrificing pricing power. Anthropic and Google publish at least indicative enterprise pricing.

3. API model proliferation makes selection decisions unnecessarily complex

OpenAI’s API spans the current GPT-5.x family (5.5, 5.4, mini, nano) alongside numerous still-available legacy models (GPT-4.1, GPT-4o, o-series). The performance/cost tradeoffs between flagship, mini, nano, and the older reasoning models are non-obvious. OpenAI’s own model selection guidance is thorough but requires significant reading. A model recommendation matrix (a simple “use X for Y task at Z cost”) embedded in the pricing page would reduce the selection friction that leads developers to overspend on premium models for commodity tasks.

Key takeaways

1. The free tier is a distribution asset, not a cost center. ChatGPT’s 300M weekly users represent the most efficient AI customer acquisition machine ever built. For consumer AI products, a genuinely useful free tier creates compounding organic growth that paid acquisition can’t match. See PLG strategy for how to structure this.

2. Price cutting is a market expansion strategy, not a margin concession. OpenAI’s 98% API price reduction since 2023 unlocked entirely new developer segments and use cases. The Batch API and GPT-4o mini didn’t cannibalize existing revenue — they created new revenue from customers who couldn’t afford prior prices. Usage-based pricing at commodity prices, with premium tiers for power users, is the sustainable AI API model.

3. Anchor pricing compounds over time. The $20 ChatGPT Plus price anchored the consumer AI subscription market. Every competitor launched at or near that price. Being first to establish a price anchor is worth more than small revenue optimization on early cohorts. See how AI companies shift from per-user licenses.

4. Reasoning models require separate pricing logic. The o-series models cannot be priced on the same per-token axis as GPT-4o without significant premium. The reason: they consume 10–100× more compute per query via extended thinking chains. Any AI company offering reasoning capabilities needs a distinct pricing tier — not just a higher price for the same billing unit. See outcome-based pricing patterns.

5. Default billing safeguards are a product feature. The absence of default API spend caps has caused documented cases of $10,000+ surprise bills from agentic loops. Every AI API product should ship with default spending alerts and configurable hard caps. This is a developer trust issue, not just a billing UX issue.

UBP implications

1. Model proliferation requires multi-dimensional pricing architecture. OpenAI’s matrix of efficiency tiers (GPT-5.x nano/mini/standard/flagship) plus reduced cached-input rates and a Batch discount creates a pricing surface that doesn’t fit any single dimension. For product teams building on OpenAI, this means usage aggregation across model types must be tracked separately to understand true cost structure. A call to GPT-5.5 has roughly 25× the input unit cost of the same call to GPT-5.4 nano — treating tokens as a uniform billing unit produces misleading cost models.

2. The Batch API creates a two-class market that changes AI product economics. The 50% Batch API discount means real-time (full price) vs. background (half price) is now a structural product architecture decision, not just a performance question. Teams should design their AI systems to route latency-tolerant tasks (batch embeddings, classification jobs, document processing) through the Batch API automatically. This usage-based billing optimization can halve infrastructure costs for workloads that don’t need sub-second responses.

3. Cost deflation makes commitment-based pricing dangerous for buyers. OpenAI has cut API prices repeatedly without notice. Any developer who signs a multi-year commitment at today’s rack rates risks being locked into above-market prices within 12 months. Until OpenAI offers volume tiers or commit discounts that reset automatically with price decreases, buyers should avoid long-term API spend commitments and favor flexible usage-based billing.

Sources

• OpenAI API pricing page (accessed 2026-05-30)
• OpenAI developer docs — API pricing (accessed 2026-05-30)
• OpenAI ChatGPT pricing page (accessed 2026-05-30)
• OpenAI platform changelog (accessed 2026-05-30)
• OpenAI blog — Introducing GPT-4.1 (accessed 2026-05-30)
• OpenAI blog — Introducing ChatGPT Pro (accessed 2026-05-30)
• OpenAI platform models documentation (accessed 2026-05-30)

Bottom line

OpenAI built the most valuable consumer AI franchise in history on a freemium subscription ($20 Plus, now flanked by an $8 Go tier and $100/$200 Pro tiers) anchored by a free product that 300M people use every week — while simultaneously running the world’s most important AI API at prices that have fallen dramatically since 2023, now on the GPT-5.x generation. The pricing architecture is not elegant by design: it grew by necessity as each new model and tier was added. The result is real complexity (many models, opaque enterprise pricing, dangerous default billing limits) that creates friction for developers and procurement teams. But the strategic wins outweigh the UX gaps: ChatGPT owns consumer mind share, the Plus price anchored the market, and the Batch API / prompt caching system gives sophisticated developers real cost leverage. The next frontier for OpenAI pricing is likely volume tiers and default budget controls — the two gaps that matter most for enterprise and developer trust.

Browse the full pricing blueprint to compare OpenAI against Anthropic, DeepSeek, and other AI platforms.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Current Pricing Snapshot

May 2026

Live capture of the API, ChatGPT, and Enterprise pricing surfaces used as the basis for the current Facts in this analysis.

GPT-5.x Flagship Line Live — GPT-5.5 $5/$30, GPT-5.4 $2.50/$15

May 2026

The API pricing page led with the GPT-5.x flagship line: GPT-5.5 at $5/1M input and $30/1M output (cached input $0.50), GPT-5.4 at $2.50/$15 (cached $0.25), and GPT-5.4 mini at $0.75 input — selectable across Standard, Batch (50% off), and Data-residency processing modes. Source: Wayback snapshot of openai.com/api/pricing.

o3 and o4-mini Launched

Apr 2025

OpenAI launched o3 ($10/1M input, $40/1M output) and o4-mini ($1.10/$4.40), both with 200K context and tool use (web search, code execution). o4-mini replaced o3-mini as the affordable reasoning model.

GPT-4.1 Family Launched — 1M Context, Focused on Coding

Apr 2025

OpenAI launched GPT-4.1 ($2/1M input, $8/1M output), GPT-4.1 mini ($0.40/$1.60), and GPT-4.1 nano ($0.10/$0.40). All support 1M token context windows. Positioned for agentic coding workflows via the API only — not in ChatGPT consumer products. The May 2025 API pricing page listed the GPT-4.1 family (with cached-input rates from $0.025/1M) alongside reasoning models o3 ($10/$40) and o4-mini ($1.10/$4.40), plus a 50% Batch API discount. Source: Wayback snapshot of openai.com/api/pricing.

ChatGPT Pro Launched at $200/month

Dec 2024

OpenAI launched ChatGPT Pro at $200/month, providing unlimited access to o1 Pro mode (extended compute reasoning), o1, o1-mini, GPT-4o, and Advanced Voice Mode. Positioned as a direct competitor to Perplexity Max.

GPT-4o mini Launched — Frontier Performance at Commodity Price

Jul 2024

GPT-4o mini launched at $0.15/1M input, $0.60/1M output — 97× cheaper than original GPT-4, competitive with early GPT-4 on many benchmarks. Replaced GPT-3.5 Turbo as the default affordable model. By September 2024 the ChatGPT pricing page listed GPT-4o and GPT-4o mini access across the Free and Plus tiers. Source: Wayback snapshot of openai.com/chatgpt/pricing.

GPT-4o Launched — 50% Cheaper than GPT-4 Turbo

May 2024

GPT-4o launched at $5/1M input, $15/1M output (later reduced to $2.50/$10 in August 2024). Multimodal (text, audio, vision) and significantly faster than GPT-4 Turbo.

ChatGPT Four-Tier Lineup — Free / Plus $20 / Team $25 / Enterprise

Jan 2024

The ChatGPT pricing page showed the four-tier consumer/business lineup: Free $0, Plus $20/mo, Team $25/seat (billed annually; $30 billed monthly), and Enterprise (Contact Sales). Team added shared GPTs, an admin console, and higher GPT-4 caps. Source: Wayback snapshot of openai.com/chatgpt/pricing.

GPT-4 Turbo Launched — 10× Cheaper

Nov 2023

GPT-4 Turbo (gpt-4-turbo-preview) launched at $10/1M input, $30/1M output — a 5× input and 2× output price cut from original GPT-4. Also introduced 128K context window. Announced at OpenAI DevDay.

GPT-4 API Launched — $60/1M Output Tokens

Mar 2023

GPT-4 API launched at $30/1K input and $60/1K output (per 1M: $30 input, $60 output for 8K context; $60/$120 for 32K context). Most expensive widely-available model at launch.

ChatGPT Plus Launched at $20/month

Feb 2023

OpenAI introduced ChatGPT Plus at $20/month, offering priority access and faster response times during peak hours. First consumer revenue stream.

DALL·E Image Models Added to API Pricing

Feb 2023

The API pricing page added a dedicated "Image models" section: 1024×1024 at $0.020/image, 512×512 at $0.018, 256×256 at $0.016, alongside the reduced GPT-3 base-model prices held from the September 2022 cut. Source: Wayback snapshot of openai.com/api/pricing.

ChatGPT Launched — Free

Nov 2022

ChatGPT launched publicly as a free product powered by GPT-3.5 Turbo. One million users in 5 days; 100M in 60 days. No paid plan at launch.

GPT-3 API Price Cut — Davinci $0.06 → $0.02/1K

Sep 2022

On September 1, 2022 OpenAI cut GPT-3 base-model prices roughly 3×: Davinci $0.0600 → $0.0200/1K, Curie $0.0060 → $0.0020/1K, Babbage $0.0012 → $0.0005/1K, Ada $0.0008 → $0.0004/1K. The pricing page carried an explicit "We made our API more affordable on September 1" banner. Source: Wayback snapshot of openai.com/api/pricing.

GPT-3 Base Model Prices — Davinci $0.06/1K

Jan 2022

Public API pricing showed four GPT-3 base models: Ada $0.0008/1K, Babbage $0.0012/1K, Curie $0.0060/1K, and Davinci $0.0600/1K tokens, plus $18 in free credit usable in the first 3 months. Source: Wayback snapshot of openai.com/api/pricing.

GPT-3 API Private Beta

Jun 2020

OpenAI launched GPT-3 in private beta with usage-based pricing. Davinci model priced at $0.02 per 1,000 tokens ($20/1M). Only selected partners had access.