OpenAI API Pricing 2026: Model Rates, ChatGPT Plans, and Startup Credits
Every current OpenAI model's per-token rate, the Batch, Flex and cached-input discounts that reshape the bill, what the ChatGPT plans cost, and the two programs that give startups OpenAI credits.
OpenAI API pricing per model (September 2026)
Standard-tier prices in USD per million tokens from OpenAI's pricing page. Cached input is the rate for prompt tokens the API has already seen.
| Model | Input | Cached input | Output | Position in the lineup |
|---|---|---|---|---|
| gpt-5.6-sol | $4.00 | $0.40 | $20.00 | Newest flagship |
| gpt-5.6-terra | $2.00 | $0.20 | $12.00 | Mid-tier of the 5.6 family |
| gpt-5.6-luna | $0.20 | $0.02 | $1.20 | Small, fast 5.6 model |
| gpt-5.5 | $5.00 | $0.50 | $30.00 | Previous flagship |
| gpt-5.5-pro | $30.00 | — | $180.00 | Extended-compute reasoning |
| gpt-5.4 | $2.50 | $0.25 | $15.00 | Workhorse general model |
| gpt-5.4-mini | $0.75 | $0.075 | $4.50 | Cost-efficient general model |
| gpt-5.4-nano | $0.20 | $0.02 | $1.25 | Cheapest 5.4 model |
| gpt-5.4-pro | $30.00 | — | $180.00 | Extended-compute reasoning |
| gpt-5.2 | $1.75 | $0.175 | $14.00 | Older generation |
| gpt-5.2-pro | $21.00 | — | $168.00 | Older pro model |
| gpt-5.1 | $1.25 | $0.125 | $10.00 | Older generation |
| gpt-5 | $1.25 | $0.125 | $10.00 | Original GPT-5 |
| gpt-5-mini | $0.25 | $0.025 | $2.00 | Original small model |
| gpt-5-nano | $0.05 | $0.005 | $0.40 | Cheapest model OpenAI sells |
| gpt-5-pro | $15.00 | — | $120.00 | Original pro model |
| o3 | $2.00 | $0.50 | $8.00 | Reasoning model |
| o3-mini | $1.10 | $0.55 | $4.40 | Small reasoning model |
| o3-pro | $20.00 | — | $80.00 | Reasoning, extended compute |
| o1 | $15.00 | $7.50 | $60.00 | Legacy reasoning |
| o1-pro | $150.00 | — | $600.00 | Legacy, most expensive |
Regional data-residency endpoints carry a 10% uplift for models released on or after March 5, 2026. The pro models do not offer cached input.
Batch, Flex, Fast mode and cached input: the multipliers
OpenAI's list price is one of four rates you can pay for the same model.
- Standard. The table above. Synchronous, full speed.
- Batch API: 50% off. Submit a file of requests and get results back asynchronously, typically within 24 hours. gpt-5.6-sol drops from $4.00 to $2.00 input. Both input and output are halved.
- Flex processing: matches Batch pricing. Same discount, but through the normal API with lower priority and slower, variable latency. Good for background jobs that still need a request-response shape.
- Fast mode: 2x standard. Faster output at double the price. gpt-5.6-sol is $8.00 input in fast mode.
- Cached input: 10% of standard. Prompt tokens the API has seen recently bill at one tenth of the input rate. Long system prompts and shared context are where this pays off.
These compound. A batch job that reuses a cached prefix pays 50% on output and as little as 5% on the cached portion of input. If your workload is offline, the difference between standard and batch-plus-cache is not a discount, it is a different business.
ChatGPT plan pricing
The consumer and business ChatGPT plans are separate from API billing. These are the plan prices as we understand them; the ChatGPT pricing page was not reachable when we last verified, so confirm on openai.com before relying on a figure.
| Plan | Price | For | Notes |
|---|---|---|---|
| Free | $0 | Anyone | Access to the default model with usage limits |
| Go | Low-cost tier | Individuals in select markets | More usage than Free at a price below Plus |
| Plus | $20/month | Individuals | Higher limits, access to newer models, Codex included |
| Pro | $200/month | Power users | Highest limits, pro reasoning models, extended Codex usage |
| Business (formerly Team) | Per seat, monthly or annual | Teams | Shared workspace, admin controls, no training on your data by default |
| Enterprise | Custom | Larger organizations | SSO, SCIM, compliance, dedicated support |
| Edu | Custom | Universities | Campus-wide access at education pricing |
Plan subscriptions do not include API credit. If you build a product on OpenAI, you pay per token regardless of which ChatGPT plan you hold.
OpenAI Codex pricing
"OpenAI codex pricing" now means the Codex coding agent (CLI, IDE and cloud), not the retired 2021 code models. Codex is bundled with ChatGPT Plus, Pro and Business plans with usage limits that scale with the plan, so most individual developers do not pay separately for it. When you run Codex against an API key instead, usage bills at the standard per-token rates of whichever model Codex is using, with the same cached-input discount.
Budget for it the way you would for any agent: a session that reads a large codebase, runs tests and iterates can consume tens of thousands of input tokens per turn, most of them cached after the first pass. On gpt-5.6-sol that is a few dollars per working session at list price; on a subscription it is a slice of a fixed allowance.
The practical comparison for a startup is the same one we make for Claude Code: interactive use by a human is cheapest on a subscription; automated use in CI or inside your product belongs on the API, ideally on credits.
OpenAI API cost: three worked examples
Real arithmetic from the table above. Swap in your own token counts.
| Workload | Model | Assumptions | Monthly cost (standard) | Optimized |
|---|---|---|---|---|
| Support ticket triage, 10,000 tickets/month | gpt-5.4-nano | 3,000 input, 700 output tokens per ticket | Input 30M × $0.20 = $6 Output 7M × $1.25 = $8.75 $14.75 |
Batch: $7.38 |
| RAG chatbot, 100,000 queries/month | gpt-5.4 | 5,000 input (4,000 shared context + 1,000 query), 500 output per query | Input 500M × $2.50 = $1,250 Output 50M × $15 = $750 $2,000 |
Cache the 4,000-token prefix: 400M × $0.25 = $100, plus 100M × $2.50 = $250, plus $750 output = $1,100 |
| Agentic coding, one developer, 20 sessions/month | gpt-5.6-sol | 50,000 input, 15,000 output tokens per session | Input 1M × $4 = $4 Output 0.3M × $20 = $6 $10 |
With 80% of input cached: $0.80 + $0.80 + $6 = $7.60 |
Compare the RAG row with the same workload on Claude Sonnet 5: $1,500 standard, $780 with caching. Model choice matters, but caching matters more on both platforms.
For founders
How startups get OpenAI API credits free
There are two programs and they are not the same size.
OpenAI for Startups is OpenAI's direct program. Our listing shows up to $5,000 in API credits for eligible early-stage companies, applied through openai.com/startups. It is worth doing because it is quick, but $5,000 at GPT-5.4 rates is a few months of a modest production workload, not a runway.
The fine print
Microsoft for Startups is the larger route. The Founders Hub provides up to $150,000 in Azure credits on our listing, and Azure credits pay for OpenAI models served through Azure OpenAI. Same models, same per-token shape, billed to a credit pool an order of magnitude bigger. If you are serious about building on GPT models, this is the program to structure around. VC-backed startups can access the investor offer tier for the higher amounts.
Founders who use both typically put development and evaluation on OpenAI's own credits, where the newest models land first, and production on Azure OpenAI where the credit pool can absorb real traffic. AWS Activate credits do not cover OpenAI models; they cover Claude and others on Bedrock.
OpenAI for Startups
Up to $5,000
- $2,500 in API credits via Ramp's startup rewards path
- Up to $5,000+ via OpenAI's 200+ VC and accelerator partners (a16z, Sequoia, Kleiner Perkins, Y Combinator, and more)
- Higher rate limits and usage tier upgrades
Step by step: setting up OpenAI on startup credits
Apply to both programs on the same day. Neither blocks the other.
-
1
Create an OpenAI organization on your company domain
Sign up on the OpenAI platform with a work email, verify the organization, and set a monthly usage limit before anyone ships code against it. -
2
Apply to OpenAI for Startups
Submit at openai.com/startups. Eligibility, what the credits cover and how long approval takes are on our OpenAI startup program page. -
3
Apply to Microsoft for Startups Founders Hub
The self-serve tier needs no investor. Verify up to the higher tiers with institutional funding. Our Microsoft for Startups page covers the tiers and rejection reasons. -
4
Deploy the same models through Azure OpenAI for production
Keep your prompts and eval harness portable between the OpenAI API and Azure OpenAI so you can move traffic to whichever credit pool has balance left. -
5
Turn on caching and move offline work to Batch
Cached input at 10% and Batch at 50% off are the two levers that make the credits last. Structure prompts with the shared context first so the cache hits. -
6
Model post-credit cost before the credits run out
Use the worked examples above with your own token counts. Standard pricing applies the day the balance hits zero.
Questions founders ask
How much does the OpenAI API cost?
What is the cheapest OpenAI model?
Is the OpenAI API free?
What is the difference between ChatGPT Plus and the OpenAI API?
How do OpenAI Batch and Flex pricing work?
How do I get OpenAI API credits for my startup?
How much does OpenAI Codex cost?
Related guides
Claude API pricing
Anthropic's per-token rates and worked examples for comparison.
Claude pricing
Claude plans versus ChatGPT plans.
Claude Code pricing
The Codex alternative and what it costs.
OpenAI startup program
Eligibility and how to apply for OpenAI for Startups.
Microsoft for Startups
Azure credits that cover OpenAI models.
How to get an Anthropic API key
If you are comparing providers.
