Estimate the exact monthly cost of every OpenAI model — GPT-5.6 Sol / Terra / Luna, GPT-5.5, GPT-5.4, GPT-5, GPT-4.1, GPT-4o and the o-series reasoners. Includes Standard, Batch, Flex and Fast (priority) pricing, plus image, audio, video, embedding and tool rates.
Rates verified against OpenAI’s official pricing page on .
OpenAI cost estimator
Pick your unit, dial in a typical request size, and instantly see the per-call cost and monthly total across every current OpenAI model — sorted from cheapest to most expensive by default.
Calculate by
× 30 = 15,000 /month
GPT-5 Nano Cheapest 5-series tier; strong replacement for GPT-4.1 nano. | GPT-5 | Bulk classification | Very fast | $0.0500 | $0.4000 | $0.0003 | $3.75 |
GPT-4.1 Nano Cheapest long-context model in the 4.1 line. | GPT-4.1 | Cheap long-context | Very fast | $0.1000 | $0.4000 | $0.0003 | $4.50 |
GPT-4o Mini Still a very cheap workhorse — but GPT-5 mini beats it on most benchmarks. | GPT-4o | Legacy fast chat | Very fast | $0.1500 | $0.6000 | $0.0005 | $6.75 |
GPT-5.6 Luna Cheapest 5.6-series model — pick this for high-volume, low-latency work. | GPT-5.6 | High-volume tasks | Very fast | $0.2000 | $1.20 | $0.0008 | $12.00 |
GPT-5.4 Nano Fastest and cheapest 5.4 tier — great for classifiers and simple structured outputs. | GPT-5.4 | Classifiers | Very fast | $0.2000 | $1.25 | $0.0008 | $12.38 |
GPT-4.1 Mini Fast, long-context mini — good default for RAG. | GPT-4.1 | Long-context RAG | Very fast | $0.4000 | $1.60 | $0.0012 | $18.00 |
GPT-5 Mini Chatty, cheap and fast — great replacement for GPT-4o mini. | GPT-5 | Cheap chat | Very fast | $0.2500 | $2.00 | $0.0013 | $18.75 |
GPT-5.4 Mini Sub-second first-token times at a fraction of the flagship price. | GPT-5.4 | Low-latency chat | Very fast | $0.7500 | $4.50 | $0.0030 | $45.00 |
o3 Mini Cheap reasoning tier for straightforward multi-step problems. | o-series | Cheap reasoning | Fast | $1.10 | $4.40 | $0.0033 | $49.50 |
o4 Mini Latest cheap reasoning tier — usually the right default for agent loops. | o-series | Agent default | Fast | $1.10 | $4.40 | $0.0033 | $49.50 |
GPT-4.1 1M-token context; great for whole-repo or long-document workloads. | GPT-4.1 | Long-context RAG | Fast | $2.00 | $8.00 | $0.0060 | $90.00 |
o3 The pragmatic reasoning default — much cheaper than o1 with similar quality. | o-series | Agent loops | Medium | $2.00 | $8.00 | $0.0060 | $90.00 |
GPT-5.1 Intermediate release — same rates as GPT-5. | GPT-5.1 | General chat | Fast | $1.25 | $10.00 | $0.0063 | $93.75 |
GPT-5 The original GPT-5 — still a competitive baseline. | GPT-5 | Bulk classifiers | Fast | $1.25 | $10.00 | $0.0063 | $93.75 |
GPT-4o Legacy multimodal flagship. Prefer GPT-5.4 or GPT-5.6 Terra for new work. | GPT-4o | Legacy multimodal | Fast | $2.50 | $10.00 | $0.0075 | $112.50 |
GPT-5.6 Terra The everyday default — best price/quality ratio in the GPT-5.6 line. | GPT-5.6 | Everyday production | Fast | $2.00 | $12.00 | $0.0080 | $120.00 |
GPT-5.2 Older mid-tier — usually cheaper to move to GPT-5.4 or GPT-5.6 Terra. | GPT-5.2 | Legacy chat | Fast | $1.75 | $14.00 | $0.0088 | $131.25 |
GPT-5.4 The workhorse — the price/quality sweet spot for most production traffic. | GPT-5.4 | Production default | Fast | $2.50 | $15.00 | $0.0100 | $150.00 |
GPT-5.6 Sol Flagship reasoning + coding model. Long-context tier kicks in above 272K. | GPT-5.6 | Hard reasoning | Medium | $5.00 | $30.00 | $0.0200 | $300.00 |
GPT-5.5 Previous flagship — still competitive on price and quality. | GPT-5.5 | Migration workloads | Medium | $5.00 | $30.00 | $0.0200 | $300.00 |
o1 Original chain-of-thought reasoner — expensive but unbeaten on hard math/code. | o-series | Hard reasoning | Slow | $15.00 | $60.00 | $0.0450 | $675.00 |
o3 Pro Long-thinking o3 variant — useful when o3 fails on a hard problem. | o-series | Long thinking | Slow | $20.00 | $80.00 | $0.0600 | $900.00 |
GPT-5 Pro Extended-thinking tier of GPT-5. | GPT-5 | Deep reasoning | Slow | $15.00 | $120.00 | $0.0750 | $1,125 |
GPT-5.2 Pro Older pro-tier reasoning model. | GPT-5.2 | Legacy reasoning | Slow | $21.00 | $168.00 | $0.1050 | $1,575 |
GPT-5.5 Pro Deep-reasoning tier for hardest problems in code and analysis. | GPT-5.5 | Deep analysis | Slow | $30.00 | $180.00 | $0.1200 | $1,800 |
GPT-5.4 Pro Extended-thinking variant for hard math/code — 12× cost of vanilla 5.4. | GPT-5.4 | Hard math / code | Slow | $30.00 | $180.00 | $0.1200 | $1,800 |
o1 Pro Highest-effort o1 tier — use sparingly, only for problems o1 fails. | o-series | Extreme reasoning | Slow | $150.00 | $600.00 | $0.4500 | $6,750 |
OpenAI model families
OpenAI now ships six active text-model families plus the o-series reasoners. Here’s the one-line brief for each, the workloads they were built for, and the price range you’ll pay.
The newest flagship family (Sol / Terra / Luna).
Best for: Any new project — Sol for hard problems, Terra for the default, Luna for high-volume tasks.
Previous flagship — competitive on price and quality.
Best for: Migration targets from GPT-4-Turbo / GPT-4o that don’t yet need 5.6.
The price/quality sweet spot for most production traffic.
Best for: Chatbots, summarisation, structured extraction, coding assistants.
Original GPT-5 line — great cheap-ish workhorses.
Best for: High-volume classifiers and lightweight chat.
1M-token context specialist.
Best for: Whole-repo code assistants, book-length summarisation, large-document RAG.
Legacy multimodal — kept alive for backwards compatibility.
Best for: Existing apps not yet ready to migrate. Prefer GPT-5.4 for new work.
Chain-of-thought reasoners for hard math, code and planning.
Best for: Multi-step agents, competitive programming, formal proofs, complex analysis.
OpenAI service tiers
OpenAI meters the same models at four different rates depending on how much latency you can tolerate. Batch and Flex are 50% off, Fast (formerly Priority) is 2× the standard rate for a guaranteed low-latency queue.
| Model | Input | Cached input | Output | Input (long ctx) | Cached (long) | Output (long) |
|---|---|---|---|---|---|---|
GPT-5.6 Sol GPT-5.6 | $5.00 | $0.500 | $30.00 | $10.00 | $1.00 | $45.00 |
GPT-5.6 Terra GPT-5.6 | $2.00 | $0.200 | $12.00 | $4.00 | $0.400 | $18.00 |
GPT-5.6 Luna GPT-5.6 | $0.200 | $0.020 | $1.20 | $0.400 | $0.040 | $1.80 |
GPT-5.5 GPT-5.5 | $5.00 | $0.500 | $30.00 | $10.00 | $1.00 | $45.00 |
GPT-5.5 Pro GPT-5.5 | $30.00 | — | $180.00 | — | — | — |
GPT-5.4 GPT-5.4 | $2.50 | $0.250 | $15.00 | $5.00 | $0.500 | $22.50 |
GPT-5.4 Mini GPT-5.4 | $0.750 | $0.075 | $4.50 | — | — | — |
GPT-5.4 Nano GPT-5.4 | $0.200 | $0.020 | $1.25 | — | — | — |
GPT-5.4 Pro GPT-5.4 | $30.00 | — | $180.00 | $60.00 | — | $270.00 |
GPT-5.2 GPT-5.2 | $1.75 | $0.175 | $14.00 | — | — | — |
GPT-5.2 Pro GPT-5.2 | $21.00 | — | $168.00 | — | — | — |
GPT-5.1 GPT-5.1 | $1.25 | $0.125 | $10.00 | — | — | — |
GPT-5 GPT-5 | $1.25 | $0.125 | $10.00 | — | — | — |
GPT-5 Mini GPT-5 | $0.250 | $0.025 | $2.00 | — | — | — |
GPT-5 Nano GPT-5 | $0.050 | $0.005 | $0.400 | — | — | — |
GPT-5 Pro GPT-5 | $15.00 | — | $120.00 | — | — | — |
GPT-4.1 GPT-4.1 | $2.00 | $0.500 | $8.00 | — | — | — |
GPT-4.1 Mini GPT-4.1 | $0.400 | $0.100 | $1.60 | — | — | — |
GPT-4.1 Nano GPT-4.1 | $0.100 | $0.025 | $0.400 | — | — | — |
GPT-4o GPT-4o | $2.50 | $1.25 | $10.00 | — | — | — |
GPT-4o Mini GPT-4o | $0.150 | $0.075 | $0.600 | — | — | — |
o1 o-series | $15.00 | $7.50 | $60.00 | — | — | — |
o1 Pro o-series | $150.00 | — | $600.00 | — | — | — |
o3 o-series | $2.00 | $0.500 | $8.00 | — | — | — |
o3 Pro o-series | $20.00 | — | $80.00 | — | — | — |
o3 Mini o-series | $1.10 | $0.550 | $4.40 | — | — | — |
o4 Mini o-series | $1.10 | $0.275 | $4.40 | — | — | — |
Beyond text
Text tokens are only part of the bill. Realtime voice, image generation with gpt-image-2, Sora 2 video, transcription with gpt-transcribe, embeddings and every hosted tool (web search, code interpreter, file search) has its own rate card.
Image generation
Priced per 1M tokens for both text and image content. Image output tokens are the dominant cost for anything above a thumbnail.
| Model | Text input | Cached text | Image input | Cached image | Image output |
|---|---|---|---|---|---|
gpt-image-2 Newest image generation & editing model. | $5.00 | $1.25 | $8.00 | $2.00 | $30.00 |
gpt-image-1.5 Higher-fidelity 1.5 tier with text-in-image awareness. | $5.00 | $1.25 | $8.00 | $2.00 | $32.00 |
gpt-image-1 mini Cheapest gpt-image tier — great for thumbnails and bulk work. | $2.00 | $0.20 | $2.50 | $0.25 | $8.00 |
gpt-image-1 Original gpt-image release — usually superseded by gpt-image-2. | $5.00 | $1.25 | $10.00 | $2.50 | $40.00 |
chatgpt-image-latest Same model powering image generation inside ChatGPT. | $5.00 | $1.25 | $8.00 | $2.00 | $32.00 |
Video generation
Billed per second of finished video, not per token. Higher resolutions cost more; portrait and landscape are the same rate.
| Model | Resolution | Sizes | Price / second |
|---|---|---|---|
Sora 2 Standard 720p Sora 2 video generation. | 720p | 720×1280 / 1280×720 | $0.10 |
Sora 2 Pro (720p) Pro tier at 720p — faster iteration. | 720p | 720×1280 / 1280×720 | $0.30 |
Sora 2 Pro (1024p) | 1024p | 1024×1792 / 1792×1024 | $0.50 |
Sora 2 Pro (1080p) | 1080p | 1080×1920 / 1920×1080 | $0.70 |
Realtime & voice
Realtime models bill separately for audio, text and image tokens on the same request. TTS models like tts-1 are billed per 1M characters.
| Model | Audio in | Audio cached | Audio out | Text in | Text out |
|---|---|---|---|---|---|
gpt-realtime-2.1 Newest realtime voice + text + vision model. | $32.00 | $0.40 | $64.00 | $4.00 | $24.00 |
gpt-realtime-2.1 mini Cheapest realtime tier — 3× cheaper than the flagship. | $10.00 | $0.30 | $20.00 | $0.60 | $2.40 |
gpt-realtime-2 | $32.00 | $0.40 | $64.00 | $4.00 | $24.00 |
gpt-realtime | $32.00 | $0.40 | $64.00 | $4.00 | $16.00 |
gpt-audio-1.5 | $32.00 | — | $64.00 | $2.50 | $10.00 |
gpt-audio-mini | $10.00 | — | $20.00 | $0.60 | $2.40 |
gpt-4o-mini-tts | — | — | $12.00 | $0.60 | — |
| Model | Unit | Price |
|---|---|---|
tts-1 $15.00 per 1M characters (input). | per 1M characters (input) | $15.00 |
tts-1-hd $30.00 per 1M characters (input). Higher-fidelity voices. | per 1M characters (input) | $30.00 |
gpt-realtime-translate Live audio translation, billed per minute. | per minute of audio | $0.0340 |
gpt-live-transcribe Live streaming ASR — billed per minute. | per minute of audio | $0.0170 |
gpt-realtime-whisper | per minute of audio | $0.0170 |
Transcription
Modern OpenAI ASR is per-minute; the older gpt-4o family also has per-token billing so long transcripts and diarised runs are cheaper to reason with downstream.
| Model | Input tokens /1M | Output tokens /1M | Per minute |
|---|---|---|---|
gpt-transcribe Latest & cheapest ASR — $0.0045 per minute. | — | — | $0.0045 |
gpt-4o-transcribe Token-priced + estimated $0.006/minute. | $2.50 | $10.00 | $0.0060 |
gpt-4o-mini-transcribe | $1.25 | $5.00 | $0.0030 |
gpt-4o-transcribe-diarize Adds speaker diarization on top of gpt-4o-transcribe. | $2.50 | $10.00 | $0.0060 |
Whisper The original OpenAI ASR — kept for legacy pipelines. | — | — | $0.0060 |
Embeddings
Embeddings are still the cheapest part of the OpenAI stack. Use text-embedding-3-small for most RAG; upgrade to text-embedding-3-large only when recall matters.
| Model | Input /1M tokens |
|---|---|
text-embedding-3-small Cheap general-purpose embeddings — $0.02 per 1M tokens. | $0.020 |
text-embedding-3-large Higher-quality embeddings for RAG at scale. | $0.130 |
text-embedding-ada-002 Legacy — prefer text-embedding-3-small unless downstream code depends on ada. | $0.100 |
Hosted tools
Hosted tools carry a per-call or storage cost on top of the model tokens they consume.
| Tool | Pricing |
|---|---|
Web search (all models) Applies to all GPT-5.x, o-series and GPT-4.1 models. | $10 / 1k calls + search content tokens billed at model rate |
Image web search | $10 / 1k calls + search content tokens billed at model rate |
Web search preview (reasoning models) | $10 / 1k calls + search content tokens billed at model rate |
Web search preview (non-reasoning) | $25 / 1k calls (search content tokens free) |
Containers (Shell & Code Interpreter) 5-minute minimum per session; billed by the minute after. | 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 (first 1 GB free) |
File search tool call Responses API only. | $2.50 / 1k calls |
Agent Kit / ChatKit uploads | $0.10 / GB-day (first 1 GB free per account per month) |
Cost optimisation
Eight compounding levers we’ve seen shave OpenAI bills in half — in rough order of impact per hour of engineering work.
OpenAI’s Batch API returns completions within 24 hours at 50% off list price. Nightly summarisation, evals, document enrichment, dataset labelling, backfills, embeddings recomputes — anything that doesn’t need to render in a chat UI belongs on Batch.
OpenAI automatically caches prompt prefixes ≥1024 tokens for ~5 minutes and replays them at 10–20% of the standard input rate (0.5% on GPT-4o, 10% on GPT-5.x). Put the stable system message, retrieved context and tool schemas at the start; put user turns at the end. RAG workloads see 40–70% input-cost reductions with no code changes.
The GPT-5.6 line has a 25× spread from Luna to Sol; GPT-5.4 has a 12× spread from Nano to Pro. Most production traffic (classification, extraction, summarisation) works fine on a Mini or Nano tier. Reserve Sol / Pro for the 5–10% of requests that actually need frontier quality — route those explicitly.
Output tokens cost 4–6× input tokens on every OpenAI model. A conversational answer might be 400 tokens; the same answer as { "verdict": "yes", "reasons": [...] } is 40. Combine this with response_format: { type: "json_schema" } and you get both cheaper and more parseable outputs.
The o-series (o1, o3, o3-pro) spends its money on invisible reasoning tokens — o1 emits 5–20K reasoning tokens on a single hard problem at $60/M output. For anything that isn’t multi-step math, code, or planning, GPT-5.6 Terra or GPT-5.4 will match quality at ~10% of the cost.
Fast mode doubles the standard rate for a priority queue. If your P95 latency requirement is above ~2s (most agents, most chat apps), the Standard tier already meets it. Only pay the 2× premium when you have a documented, user-visible latency SLO.
Bugs in prompt templates or agent loops can produce runaway completions that burn thousands of dollars in output tokens overnight. Set max_tokens to the largest sensible value for that endpoint (200 for a classifier, 800 for a chat reply, 4000 for a report) and return an error rather than a truncated reply — that error is your alarm bell.
Set a hard monthly limit and a soft alert at 50% and 80% inside the OpenAI dashboard. Combine with per-key tags (production, staging, evals) so a runaway staging job can’t take down your production budget.
Methodology & sources
Every rate on this page is copied from OpenAI’s official pricing documentation and verified on a rolling monthly cadence. Last verified: .
The number in the “Total cost” column is a deterministic function of four inputs — model, input tokens per call, output tokens per call, and calls per day:
input_cost = (input_tokens / 1,000,000) × input_rate
output_cost = (output_tokens / 1,000,000) × output_rate
cost_per_call = input_cost + output_cost
monthly_bill = cost_per_call × requests_per_day × 30All prices on this page are transcribed directly from OpenAI’s official developer documentation. Where OpenAI publishes multiple service tiers (Standard, Batch, Flex, Fast) or context-length tiers (short vs long), we mirror both.
developers.openai.com/api/docs/pricingLast verified July 31, 2026. If a number here disagrees with OpenAI’s page, treat the linked source as authoritative and email us so we can update.
The full multi-provider calculator ranks GPT-5.6 against Claude Opus 5, Gemini 3.1 Pro, DeepSeek V4, Grok 4.5, Qwen 3.7 and 40+ others at live prices.