OpenAI's API is pay-as-you-go, billed per million tokens with no subscription. OpenAI API pricing starts at $0.10 input per 1M tokens for gpt-4.1-nano; the flagship gpt-5.5 costs $5.00 input and $30.00 output per 1M tokens. This guide lists what every model and operation costs.
The OpenAI API is pay-as-you-go, billed per million tokens with no subscription fee. The flagship gpt-5.5 costs $5.00 input and $30.00 output per 1M tokens; gpt-5.4 is half that at $2.50/$15.00 and is the recommended production model for most teams. The cheapest capable model, gpt-4.1-nano, starts at $0.10 input and $0.40 output. The Batch API cuts all token costs by 50%, and prompt caching cuts repeated input by up to 90%. New accounts receive $5 in free credits with no credit card required.
OpenAI API billing is usage-based and metered per million tokens. Every call generates an input token count and an output token count, and you pay for both at separate rates. Output tokens are consistently 4 to 8 times more expensive than input tokens. There is no monthly fee, no per-request charge, and no seat licence. The minimum prepaid top-up is $5, and credits expire after a set period.
No. ChatGPT plans (Plus, Pro, Business, Enterprise) are billed per seat per month and do not include OpenAI API credits. API usage is metered separately at platform.openai.com and must be funded with its own prepaid balance.
OpenAI API pricing for large language models is billed per million tokens across three families: the gpt-5.x flagship series, the gpt-4.1 long-context series, and the o-series reasoning models. The gpt-5.5 and gpt-5.4 families carry a 1M-token context window; prompts above 270K input tokens on gpt-5.5 and gpt-5.4 carry a higher rate for the full session. Reasoning models (o3, o4-mini) bill hidden thinking tokens at the output rate, making real costs significantly higher than the listed rate suggests.
| Model | Input / 1M | Cached Input / 1M | Output / 1M | Context |
| GPT-5.x flagship | ||||
| gpt-5.5 | $5.00 | $0.50 | $30.00 | 1M (<270K) |
| gpt-5.5-pro | $30.00 | — | $180.00 | 1M (<270K) |
| gpt-5.4 | $2.50 | $0.25 | $15.00 | 1M (<270K) |
| gpt-5.4-mini | $0.75 | $0.075 | $4.50 | 1M |
| gpt-5.4-nano | $0.20 | $0.02 | $1.25 | 1M |
| gpt-5.2 | $1.75 | $0.175 | $14.00 | 1M |
| gpt-5.1 / gpt-5 | $1.25 | $0.125 | $10.00 | 1M |
| gpt-5-mini | $0.25 | $0.025 | $2.00 | 1M |
| gpt-5-nano | $0.05 | $0.005 | $0.40 | 1M |
| GPT-4.1 series | ||||
| gpt-4.1 | $2.00 | $0.50 | $8.00 | 1M |
| gpt-4.1-mini | $0.40 | $0.10 | $1.60 | 1M |
| gpt-4.1-nano | $0.10 | $0.025 | $0.40 | 1M |
| gpt-4o | $2.50 | $1.25 | $10.00 | 128K |
| gpt-4o-mini | $0.15 | $0.075 | $0.60 | 128K |
| o-series reasoning | ||||
| o4-mini | $1.10 | $0.275 | $4.40 | — |
| o3 | $2.00 | $0.50 | $8.00 | — |
| o3-pro | $20.00 | — | $80.00 | — |
| o1 | $15.00 | $7.50 | $60.00 | — |
o-series models spend internal tokens on chain-of-thought steps that are not visible in the response but are billed at the output rate. Always set max_completion_tokens to cap worst-case spend, and monitor the reasoning_tokens field in the usage object to track actual cost per request.
OpenAI offers four processing tiers that change cost and latency. Standard is the default real-time tier at listed rates. Batch and Flex both cut all token costs 50% for workloads that can tolerate a delay, with Batch guaranteeing results within 24 hours via file upload and Flex operating per-request with variable latency. Priority delivers faster, more predictable performance at a premium above Standard. The table below shows what key models cost across all four tiers.
| Model | Standard | Batch / Flex (50% off) | Priority | Best for |
| gpt-5.5 | $5.00 in / $30.00 out | $2.50 in / $15.00 out | $12.50 in / $75.00 out | Complex agentic work |
| gpt-5.4 | $2.50 in / $15.00 out | $1.25 in / $7.50 out | $5.00 in / $30.00 out | Production workloads |
| gpt-5.4-mini | $0.75 in / $4.50 out | $0.375 in / $2.25 out | $1.50 in / $9.00 out | Budget production |
| gpt-4.1-nano | $0.10 in / $0.40 out | $0.05 in / $0.20 out | $0.20 in / $0.80 out | High-volume routing |
| o4-mini | $1.10 in / $4.40 out | $0.55 in / $2.20 out | $2.00 in / $8.00 out | Budget reasoning |
| o3 | $2.00 in / $8.00 out | $1.00 in / $4.00 out | $3.50 in / $14.00 out | Advanced reasoning |
OpenAI's multimodal operations are each priced separately on top of text token costs. AI image generation (gpt-image-2) is billed per million tokens; AI video generation (Sora) is billed per second of output; Realtime audio is billed per million audio tokens; and transcription is billed per minute or per million tokens.
gpt-image-2 is the current flagship image generation model at $8.00 image input and $30.00 image output per 1M tokens. gpt-image-1-mini is the most affordable at $2.50 input and $8.00 output. For exact per-image cost estimates use the calculator in OpenAI's image generation guide. The Batch API applies a 50% discount to image generation rates.
| Model | Modality | Input / 1M | Cached Input / 1M | Output / 1M |
| gpt-image-2 | Image | $8.00 | $2.00 | $30.00 |
| Text | $5.00 | $1.25 | — | |
| gpt-image-1.5 | Image | $8.00 | $2.00 | $32.00 |
| Text | $5.00 | $1.25 | $10.00 | |
| gpt-image-1-mini | Image | $2.50 | $0.25 | $8.00 |
| Text | $2.00 | $0.20 | — | |
| gpt-image-1 | Image | $10.00 | $2.50 | $40.00 |
| Text | $5.00 | $1.25 | — |
Sora models are billed per second of output. sora-2 costs $0.10 per second at 720p. sora-2-pro ranges from $0.30 per second at 720p up to $0.70 per second at 1080p. The Batch API halves all rates.
| Model | Size | Portrait | Landscape | Price / Sec | Best for |
| sora-2 | 720p | 720x1280 | 1280x720 | $0.10 | Short clips, prototyping |
| sora-2-pro | 720p | 720x1280 | 1280x720 | $0.30 | High-fidelity production video |
| 1024p | 1024x1792 | 1792x1024 | $0.50 | ||
| 1080p | 1080x1920 | 1920x1080 | $0.70 |
The Realtime API powers speech-to-speech voice agents and AI chatbots. gpt-realtime-2.1 bills audio at $32.00 input and $64.00 output per 1M tokens. The mini variant is $10.00/$20.00. Classic text-to-speech (tts-1) is $15.00 per 1M characters. For transcription, gpt-4o-transcribe costs approximately $0.006 per minute; gpt-4o-mini-transcribe is half that at $0.003 per minute; Whisper is $0.006 per minute billed flat.
| Model | Modality | Input / 1M | Cached Input / 1M | Output / Est. Cost |
| gpt-realtime-2.1 | Audio | $32.00 | $0.40 | $64.00 / 1M |
| Text | $4.00 | $0.40 | $24.00 / 1M | |
| Image | $5.00 | $0.50 | — | |
| gpt-realtime-2.1-mini | Audio | $10.00 | $0.30 | $20.00 / 1M |
| Text | $0.60 | $0.06 | $2.40 / 1M | |
| gpt-realtime-translate | Audio | — | — | $0.034 / min |
| tts-1 | Text in | $15.00 / 1M chars | — | — |
| tts-1-hd | Text in | $30.00 / 1M chars | — | — |
| gpt-4o-transcribe | Audio in | $2.50 / 1M tokens | — | ~$0.006 / min |
| gpt-4o-mini-transcribe | Audio in | $1.25 / 1M tokens | — | ~$0.003 / min |
| Whisper | Audio in | — | — | $0.006 / min |
Web search costs $10.00 per 1,000 calls for all models, with search content tokens billed at the chosen model's input rate. The web search preview for non-reasoning models costs $25.00 per 1,000 calls, with search content tokens free. Containers (Hosted Shell and Code Interpreter) are billed per 20-minute session. File search storage runs $0.10 per GB per day with the first 1 GB free. Embeddings for vector database and search workloads are billed for input tokens only, starting at $0.02 per 1M for text-embedding-3-small.
| Tool / Model | Details | Pricing |
| Web search | All models / reasoning / gpt-5 / o-series | $10.00 / 1k calls + search content tokens at model rates |
| Non-reasoning models (preview) | $25.00 / 1k calls + search content tokens free | |
| Containers | Hosted Shell and Code Interpreter | 1 GB $0.03, 4 GB $0.12, 16 GB $0.48, 64 GB $1.92 per 20-min session |
| File search | Storage | $0.10 / GB per day (1 GB free) |
| Tool call | $2.50 / 1k calls | |
| Embeddings | text-embedding-3-small | $0.02 / 1M tokens (input only) |
| text-embedding-3-large | $0.13 / 1M tokens (input only) | |
| text-embedding-ada-002 | $0.10 / 1M tokens (input only) |
The same 1,000 chat turns cost $9.80 on gpt-5.4 and $0.31 on gpt-4.1-nano, a 97% difference that makes model selection the single biggest cost lever available. Add the Batch API on top and those figures halve again for any non-real-time workload. The table below shows worked examples across models and operations so you can estimate your actual bill before you build.
| Scenario | Math | Cost |
| gpt-5.4 chat turn (1,500 in / 400 out) | (1,500 x $2.50 + 400 x $15.00) / 1M | ~$0.0098 |
| 1,000 such gpt-5.4 chat turns | 1,000 x $0.0098 | ~$9.80 |
| Same 1,000 turns via Batch API | $9.80 x 0.50 | ~$4.90 |
| gpt-4.1-nano chatbot turn (1,500 in / 400 out) | (1,500 x $0.10 + 400 x $0.40) / 1M | ~$0.00031 |
| o4-mini reasoning task (3K in + ~5K thinking / 500 out) | (3K x $1.10 + 5,500 x $4.40) / 1M | ~$0.0275 |
| 10-sec sora-2 video clip | 10 x $0.10 | $1.00 |
| 1-hour transcription via Whisper | 60 x $0.006 | $0.36 |
| 1,000 web search calls | 1 x $10.00 | $10.00 |
Making your first call takes three steps. New accounts receive $5 in free credits automatically with no credit card required.
prompt_tokens, completion_tokens, and for reasoning models, reasoning_tokens. Log these to track actual cost per request.gpt-4.1-nano is the cheapest capable model at $0.10 per 1M input tokens and $0.40 per 1M output tokens, with a 1M-token context window. For the newer GPT-5 architecture at low cost, gpt-5-nano is $0.05 input and $0.40 output per 1M tokens.
There is no ongoing free tier, but new accounts receive $5 in free credits with no credit card required. Credits expire after a set period after registration and work across all API services, including text, images, audio, and embeddings.
The Batch API accepts JSONL file uploads and processes requests asynchronously within 24 hours, applying a flat 50% discount on all input and output tokens across every model. It is the single largest cost lever for non-real-time workloads such as document extraction, bulk classification, content generation, and evaluations.
OpenAI automatically caches repeated prompt prefixes of roughly 1,024 tokens or longer. gpt-5.5 cached input is $0.50 per 1M (90% off), gpt-5.4 cached input is $0.25 per 1M (90% off), and gpt-4.1 cached input is $0.50 per 1M (75% off). No code changes are required. Place stable content such as the system prompt at the start of every request to maximise cache hits.
o-series models spend internal tokens on chain-of-thought reasoning steps that are not visible in the response but are billed at the output token rate. A request generating 500 visible output tokens may consume 5,000 or more reasoning tokens behind the scenes. Set max_completion_tokens to cap spend and monitor reasoning_tokens in the usage object to measure actual cost per request.
No. ChatGPT plans (Plus, Pro, Business, Enterprise) are billed per seat per month and are entirely separate from OpenAI API access. API usage must be funded with its own prepaid balance at platform.openai.com.
Standard delivers responses in real time at listed rates. Batch and Flex both cut token costs 50% for non-real-time workloads. Batch guarantees 24-hour turnaround via file upload; Flex is per-request with variable latency and possible resource unavailability. Priority provides faster, more predictable performance at a premium above Standard for latency-sensitive production workloads.
OpenAI API rate limits scale with cumulative spend across five tiers from Tier 1 ($5) through Tier 5 ($1,000). Higher tiers unlock more requests per minute and tokens per minute. Exact limits are in the Limits section of your organisation settings at platform.openai.com.
No. OpenAI API pricing on Amazon Bedrock is billed through AWS and may differ from direct OpenAI pricing. Regional processing (data residency) endpoints also carry a 10% price uplift for models released on or after March 5, 2026.
The OpenAI API suits teams that need a single platform covering text, voice, AI image generation, AI video generation, transcription, and tool use under one account. gpt-5.4 and gpt-5.4-mini are the right starting point for most production workloads at half the cost of gpt-5.5. For budget-sensitive work, gpt-4.1-nano at $0.10 input is hard to beat. The Batch API's 50% discount makes async workloads like data extraction, content pipelines, and evals very cost-efficient at scale.
In short: start on Standard pay-as-you-go to validate your use case, route simple tasks to nano or mini tiers rather than defaulting to the flagship, apply prompt caching for any repeated system prompt or shared context, and move non-real-time jobs to the Batch API once in production. OpenAI API pricing changes frequently, so confirm current rates on OpenAI's website before committing a production budget.
Check out our AI API pricing guide to compare API costs for other leading AI tools.
The pricing and product details above are sourced from OpenAI's official API pricing documentation and were accurate as of publication. Rates are subject to change. Please verify current pricing on OpenAI's website before making a purchase.