Aumiqx Technologies
AUM
Open menu

ChatGPT API Pricing 2026: What 1M Tokens Actually Cost

GPT-6 Astra, Sol, Terra and Luna priced per token, with cost examples, batch math, enterprise notes and a Claude and Gemini comparison.

Pricing|Aumiqx Team|Updated |13 min read
chatgpt api pricingopenai api pricingchatgpt enterprise pricing

ChatGPT API Pricing in 2026: The Full Model-by-Model Breakdown

Last updated: November 2026. The model line-up in this guide has changed completely since April. We originally priced GPT-4o, GPT-4.1 and the o-series. Per OpenAI's API pricing page, the current text line-up is the GPT-6 family: GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna. The older GPT-4o, GPT-4.1 and o-series prices we listed in April no longer appear on that page, so we have removed them rather than guess.

OpenAI charges per token, not per request. A token is roughly 4 characters or 0.75 words. Every call has input tokens (what you send) and output tokens (what the model generates), and output is always more expensive. Here is the current standard pricing for prompts under 272K tokens:

ModelInput (per 1M tokens)Cached inputOutput (per 1M tokens)Best For
GPT-6 Astra$10.00$1.00$50.00The hardest end-to-end work
GPT-6.1 Sol$2.00$0.10$10.00Complex coding and agentic workflows at scale
GPT-6.1 Terra$1.00$0.05$5.00Fast, efficient: visual reasoning, computer use, writing
GPT-6 Luna$0.10$0.01$0.50Focused, high-volume tasks

As of November 2026, the Batch option is 50% off, and data residency adds 10%. OpenAI also lists Flex processing (cheaper, slower, occasionally unavailable) and a Fast mode for pay-as-you-go speed. Sol and Luna prices were cut 50% on September 22, according to OpenAI's announcement (Sol from $4/$20, Luna from $0.20/$1.20).

GPT-6 Family Pricing: Which Model Is the Workhorse?

If you used GPT-4o as your default last spring, the GPT-6 family replaces it with a clearer ladder. Think of it as four rungs.

GPT-6.1 Sol: $2.00 Input / $10.00 Output

Sol is the default for coding and agent work. At $2/$10 it sits where GPT-4o used to ($2.50/$10), but it is positioned by OpenAI for complex coding and agentic workflows. This is the model to start with if you are not sure.

Rough costs on Sol:

  • A chatbot exchange (500 input tokens, 300 output tokens): about $0.004.
  • Summarizing a 10-page document (about 4,000 tokens in, 500 out): about $0.013.
  • A 1,000-word article (about 200 tokens in, 1,300 out): about $0.013.

GPT-6.1 Terra: $1.00 Input / $5.00 Output

Terra is the fast, efficient middle tier, aimed at visual reasoning, precise computer use and natural writing. It costs half of Sol. Use it for interactive features where latency and cost both matter.

GPT-6 Luna: $0.10 Input / $0.50 Output

Luna is the budget workhorse, 20x cheaper than Sol on input. It is the right default for classification, routing, extraction and simple Q&A. For example, classifying 100,000 support tickets of about 200 tokens each is about 20M input tokens, or roughly $2 of input plus a small amount of output.

GPT-6 Astra: $10.00 Input / $50.00 Output

Astra is five times the price of Sol and is OpenAI's most capable model. Use it only where Sol demonstrably fails. For many teams that will be a small slice of traffic.

For specialised needs there are also GPT-6 Rosalind Research (a life-sciences model available only to approved Research customers) and the image and voice models covered below. See OpenAI's pricing page for the full list.

What Happened to o1, o3 and the Reasoning Models?

In April this section covered o1, o3 and o3 mini. The o-series is now legacy: OpenAI has not released new o-models, o4-mini left ChatGPT in February 2026 and o3 left ChatGPT on August 26, 2026 (the API was not changed by that retirement). The current pricing page no longer lists them, so check OpenAI's docs for API availability before building on them.

Reasoning Costs Still Matter

The cost lesson from the o-series still applies to any model that thinks before answering. Internal reasoning tokens are generally billed as output tokens even when you don't see them, so a response with 500 visible tokens can bill several thousand. OpenAI's GPT-6 models offer reasoning in ChatGPT, and the API's FAQ recommends its reasoning models for complex, multi-step and STEM problems. Measure the real token usage in the usage dashboard on your own prompts before projecting costs.

When to Pay for More Capability

Escalate in this order: Luna for simple, high-volume work; Terra when quality or multimodality matters; Sol for coding and agents; Astra only for the hardest tasks. Do not pay Astra rates for summarization, extraction or classification.

ChatGPT Enterprise Pricing: What Large Organizations Actually Pay

ChatGPT Enterprise is OpenAI's top-tier plan for large organizations, and OpenAI does not publish its price. It is negotiated with sales and depends on seat count, contract length and the features you need. In April we repeated an industry estimate of $50 to $60 per user per month. We can't verify that figure, so we have removed it. Treat any number you see online as a rough anchor, not a quote.

What We Can Confirm

ItemDetails (per OpenAI's pricing page, Nov 2026)
Business planAvailable from 2 users, monthly or annual billing
EnterpriseAnnual plans, custom pricing, alternative payment options such as invoicing
Extra accessBusiness and Enterprise can purchase credits for more usage
NonprofitsUp to 75% discount on Business or Enterprise via OpenAI for Nonprofits
APIBilled separately; ChatGPT plans do not include API usage

What Enterprise Adds Over Business

Typical enterprise features include SSO and SCIM provisioning, enterprise key management, audit and analytics tooling, data residency options and support with SLAs. OpenAI also sells Scale Tier and Reserved Capacity on the API side for larger workloads. Contact sales through the ChatGPT Enterprise page for exact terms, and compare against Claude Enterprise (listed at $20 per seat plus usage at API rates) and Google's offerings.

Batch API and Cost-Saving Strategies: Cut Your OpenAI Bill

The Batch API is 50% off standard input and output prices in exchange for asynchronous processing over 24 hours. If you're not using it for eligible workloads, you are paying double.

Batch Pricing (50% Off Standard Rates)

ModelStandard InputBatch InputStandard OutputBatch Output
GPT-6 Astra$10.00$5.00$50.00$25.00
GPT-6.1 Sol$2.00$1.00$10.00$5.00
GPT-6.1 Terra$1.00$0.50$5.00$2.50
GPT-6 Luna$0.10$0.05$0.50$0.25

This is ideal for document processing, content generation at scale, data extraction and evaluation runs.

Prompt Caching

Cached input is much cheaper than fresh input: $0.10 vs $2.00 on Sol (95% off), $1.00 vs $10.00 on Astra, $0.05 vs $1.00 on Terra and $0.01 vs $0.10 on Luna. If you send the same system prompt or context prefix repeatedly, structure prompts so the shared part comes first.

Other Cost Optimization Strategies

  • Model routing. Send simple queries to Luna and escalate only when needed. A router that sends most traffic to the cheapest tier can cut bills dramatically.
  • Output token limits. Set max_output_tokens. Output costs 5x input across the GPT-6 family.
  • Flex processing for non-production or lower priority tasks.
  • Watch tool fees. The web search tool is $10 per 1,000 calls on top of token costs.

API Cost Calculator: Real-World Scenarios for Startups and Enterprises

Here are concrete calculations using November 2026 list prices. They are worked examples with assumed token counts, not benchmarks of real usage.

Scenario 1: AI Customer Support Chatbot

ParameterValue
Monthly conversations50,000
Average input per conversation800 tokens
Average output per conversation400 tokens

That is 40M input and 20M output tokens a month. On Luna: (40M x $0.10) + (20M x $0.50) = $4 + $10 = $14/month. On Terra: $40 + $100 = $140/month. On Sol: $80 + $200 = $280/month.

Scenario 2: Content Generation Pipeline

500 articles a month, 1,500 prompt tokens and 2,000 output tokens each: 0.75M input and 1M output. On Sol: $1.50 + $10.00 = $11.50/month. With the Batch API: $5.75/month.

Scenario 3: Enterprise Document Analysis

10,000 documents a month, each 8,000 tokens plus a 2,000-token system prompt, with 500 output tokens, on Sol. That is 100M input (20M of it the shared prompt) and 5M output.

  • Standard: $200 + $50 = $250/month.
  • With prompt caching on the shared 20M: $2 + $160 + $50 = $212/month.
  • With Batch only: $125/month.

Whether caching and batch discounts stack depends on your account setup, so confirm with OpenAI before budgeting for both.

Scenario 4: Reasoning-Heavy Research Tool

5,000 queries a month, 2,000 input tokens each, and an assumed 6,000 billed output tokens per query (1,000 visible plus 5,000 reasoning). That is 10M input and 30M output tokens. On Sol: $20 + $300 = $320/month. On Astra: $100 + $1,500 = $1,600/month. The same workload costs five times as much on Astra, so test whether Sol is good enough.

These calculations use published rates from OpenAI's pricing page. Actual costs vary with prompts, response variability and whether you use caching and batching.

OpenAI API vs. Claude API vs. Gemini API: Full Cost Comparison

Cost is only one factor in choosing an API provider, but it is a big one. Prices below are per 1 million tokens (input / output), as of November 2026.

TierOpenAIAnthropic (Claude)Google (Gemini)
TopGPT-6 Astra: $10 / $50Claude Fable 5.2: $8 / $40Gemini 4 Carbon (introductory $2 / $10 through Nov 30; later price not confirmed)
WorkhorseGPT-6.1 Sol: $2 / $10Opus 5.5: $4 / $20; Sonnet 5.5: $2 / $10Gemini 4 Argon (introductory $2 / $10, then $4 / $20 per Google)
FastGPT-6.1 Terra: $1 / $5Sonnet 5.5: $2 / $10Gemini 4 Hydrogen: $1.50 / $7.50
BudgetGPT-6 Luna: $0.10 / $0.50Haiku 5.5: $0.10 / $0.50 (prompts up to 100K)Gemini 4 Hydrogen has a free tier
Batch50% offDiscounted batch pricing, see docsSee Google pricing docs
Cached inputSol $0.10 (95% off)Sonnet 5.5 cache hit $0.10 (95% off)Argon cached input 95% off (introductory)

Other labs worth pricing: xAI's Grok 4.7 is $2 / $6 per million tokens per its docs, and DeepSeek's V4.1-Flash is $0.15 / $0.60 off-peak ($0.30 / $1.20 at peak).

Key Takeaways from the Comparison

At the headline tiers, prices have converged. OpenAI's Sol and Anthropic's Sonnet 5.5 are both $2 / $10, and the budget tiers (Luna and Haiku 5.5) are both $0.10 / $0.50 for short prompts. Choose on quality for your task, not on list price.

Watch the tokenizer. Anthropic notes that Claude 4.7 and later models produce about 30% more tokens for the same text, so equal list prices don't mean equal bills. Measure real usage.

Google's Hydrogen is the one to watch for free-tier prototyping, and Argon and Carbon launch pricing was introductory, so recheck before committing.

For Claude's full structure, see our Claude pricing breakdown. For consumer plans, see our ChatGPT pricing guide.

Image, Voice and Tool Pricing

Beyond text, OpenAI sells image, voice and tool APIs. We could not confirm current embedding, DALL-E or Whisper prices on the pricing page this month, so we have dropped the older numbers; check the docs if you rely on them.

Image Generation

The current image models are GPT-Image-2.6 Flare (the fastest, for high-quality everyday generation) and GPT-Image-2.6 Sunburst (for editing precision), launched on November 4, 2026. Both are priced the same: image input $8 per 1M tokens ($2 cached), image output $30 per 1M tokens, and text input $5 per 1M tokens ($1.25 cached). Images are billed per token, so cost per image depends on size and quality.

Voice and Transcription

ModelPriceUse
GPT-Live-1$0.05 per minuteExpressive voice conversations
GPT-Realtime-2.1Audio $32 in / $64 out per 1M tokens; text $4 / $24Realtime voice agents with reasoning
GPT-Realtime-2.1 miniAudio $10 in / $20 out per 1M tokens; text $0.60 / $2.40Cheaper realtime voice
GPT-Transcribe$0.0045 per minuteAsync and batch transcription
GPT-Live-Transcribe$0.017 per minuteStreaming transcription
GPT-Realtime-Translate$0.034 per minuteLive speech translation

Transcribing an hour of audio with GPT-Transcribe works out to about $0.27 (60 x $0.0045). Sora, which we covered in April, was discontinued: the app and web in April and the API in September 2026. Web search is $10 per 1,000 calls.

API Access vs. ChatGPT Plans: Which Is Right for Your Organization?

Organizations often ask whether to buy ChatGPT Business or Enterprise seats or to build internal tools on the API. The two are billed separately, because OpenAI states API usage isn't included in ChatGPT Plus, Business, Enterprise or Edu.

When to Choose ChatGPT Seats

  • Non-technical teams need AI access in a polished interface, with no custom software.
  • You need it deployed fast with admin and security controls built in.
  • Use cases are diverse: writing, analysis, brainstorming and coding in one place.

When to Choose the API

  • You're building a product: customer-facing chatbots, document pipelines and recommendations need the API.
  • Volume economics favor it. Our chatbot scenario above costs $14 to $280 a month for 50,000 conversations, depending on the model, which may be far below the cost of seats for the same volume.
  • You need customization: custom system prompts, structured outputs, function calling and integration with internal systems.
  • Batch processing at 50% off beats any seat-based plan for large offline jobs.

The Hybrid Approach

Most large organizations use both: seats for knowledge workers and API integration for high-volume workflows. Pro plans for individuals now also include monthly API credits ($100, $200 or $500 depending on the tier, per OpenAI's pricing page), which is worth knowing if you are a developer paying for both. For a broader view of AI in business workflows, explore our automation guides.

Getting Started: Credits, Rate Limits and Spending Controls

Create an account at platform.openai.com and generate an API key. In April we quoted $5 in free credits for new accounts and a table of usage-tier rate limits. We couldn't re-verify those numbers, so check the platform dashboard and OpenAI's rate-limit docs for what applies to your account.

Spending Controls

OpenAI's FAQ says you can set a monthly budget in your billing settings, after which requests stop being served. There may be a delay in enforcing the limit and you are responsible for any overage. You can also set an email notification threshold and per-project billing restrictions. Set these on day one. A misconfigured loop can burn hundreds of dollars in hours. Playground usage is billed like regular API usage.

For production applications, monitor costs through the usage dashboard, especially if you expose the API to end users.

API Pricing Trends: Where OpenAI Costs Are Heading

If you build on OpenAI's API, pricing trends affect your architecture, not just this month's bill.

The Trend: More Capability per Dollar

The clearest recent data point is OpenAI's own: on September 22 it cut GPT-6 Sol and Luna prices by 50% compared with GPT-5.6 Sol ($4/$20) and Luna ($0.20/$1.20). At the same time the top end moved up: Astra at $10/$50 is priced as a premium tier. The pattern is a widening ladder rather than a uniform price drop.

What This Means for Your Architecture

  • Don't over-optimize for today's prices. Focus on big wins (batch, routing, caching).
  • Build for model switching. Model names changed completely in six months: GPT-4.1 and o3 in April, GPT-6 in November. Keep model IDs in config.
  • Negotiate at scale. OpenAI offers Scale Tier and Reserved Capacity for larger workloads. Contact sales if you spend heavily.

Competition Keeps Pressure On

Anthropic's Sonnet 5.5 and Haiku 5.5 match OpenAI's Sol and Luna list prices, Google's Hydrogen has a free tier, and DeepSeek and others undercut on price. Open-weight models provide a floor. Our AI tools directory tracks the landscape as it evolves.

The Bottom Line: What You'll Actually Spend on OpenAI's API

The blunt summary of ChatGPT API pricing in November 2026:

The ladder: Luna at $0.10/$0.50, Terra at $1/$5, Sol at $2/$10 and Astra at $10/$50 per million tokens. Batch is 50% off, and cached input is dramatically cheaper.

For prototypes: a few dollars a month on Luna or Terra. Our 50,000-conversation chatbot example is $14 a month on Luna.

For production apps: route most traffic to the cheapest model that passes your quality bar, use Batch for anything offline, and cache shared prompts. Send only the hard cases to Sol, and Astra only when Sol fails.

For enterprises: Enterprise seats are negotiated, API is billed separately, and a hybrid is typical.

Start with the cheapest model that works and upgrade only when quality requires it. Test on your own data. For the latest numbers always check OpenAI's pricing page, and see our AI tools directory and ChatGPT consumer pricing guide.

Key Takeaways

  1. 01OpenAI's API line-up is now GPT-6 Astra ($10/$50), GPT-6.1 Sol ($2/$10), GPT-6.1 Terra ($1/$5) and GPT-6 Luna ($0.10/$0.50) per million tokens (input/output)
  2. 02GPT-4o, GPT-4.1 and the o-series no longer appear on OpenAI's pricing page; the o-series is legacy
  3. 03Batch processing is 50% off and cached input is far cheaper than fresh input (Sol: $0.10 vs $2.00)
  4. 04Sol and Luna were cut 50% on Sep 22, 2026; Sonnet 5.5 matches Sol's $2/$10, and Haiku 5.5 matches Luna's $0.10/$0.50 for short prompts
  5. 05ChatGPT Enterprise pricing is not published; API usage is billed separately from any ChatGPT plan
  6. 06Sora is discontinued; the current image models are GPT-Image-2.6 Flare and Sunburst

Frequently Asked Questions

Related Guides

Mentioned Tools

ChatgptClaudeGemini