Calculate GPT article costs here by model, word count and monthly volume, using each model’s supported processing tiers.
Catalog rates verified . See verification and correction history.
Base text-token estimate, not an invoice. Excludes reasoning beyond the word estimate, output overshoot, cache writes/storage, tools, regional surcharges and editing. Unsupported tiers use Standard and say so. Check the calculation assumptions and budget for human editing.
One 1,500-word article costs $0.00080 on GPT-5 nano, OpenAI's cheapest tracked model, and $0.3627 on GPT-5.4 Pro, its most expensive. On GPT-5 nano, Batch costs $0.00040 at these assumptions.
This is a price ranking, not a writing-quality recommendation. Check the model-selection guide and real AI-written article examples before deciding what fits your editorial workflow.
One draft, a 250-word brief, 20% prompt overhead. Change any of it in the calculator.
| Model | 500w | 1,000w | 1,500w | 2,000w | 3,000w |
|---|---|---|---|---|---|
| GPT-5 nano | $0.00028 | $0.00054 | $0.00080 | $0.00106 | $0.00158 |
| GPT-4.1 nano | $0.00030 | $0.00056 | $0.00082 | $0.00108 | $0.00160 |
| GPT-6 Luna | $0.00036 | $0.00069 | $0.00101 | $0.00134 | $0.00199 |
| GPT-4o mini | $0.00045 | $0.00084 | $0.00123 | $0.00162 | $0.00240 |
| GPT-5.6 Luna | $0.00086 | $0.00164 | $0.00242 | $0.00320 | $0.00476 |
| GPT-5.4 nano | $0.00089 | $0.00170 | $0.00252 | $0.00333 | $0.00495 |
| GPT-4.1 mini | $0.00120 | $0.00224 | $0.00328 | $0.00432 | $0.00640 |
| GPT-5 mini | $0.00140 | $0.00270 | $0.00400 | $0.00530 | $0.00790 |
| GPT-5.4 mini | $0.00322 | $0.00614 | $0.00907 | $0.0120 | $0.0178 |
| GPT-4.1 | $0.00598 | $0.0112 | $0.0164 | $0.0216 | $0.0320 |
| GPT-5.1 | $0.00699 | $0.0135 | $0.0200 | $0.0265 | $0.0395 |
| GPT-5 | $0.00699 | $0.0135 | $0.0200 | $0.0265 | $0.0395 |
| GPT-6 Sol | $0.00728 | $0.0138 | $0.0203 | $0.0268 | $0.0398 |
| GPT-4o | $0.00747 | $0.0140 | $0.0205 | $0.0270 | $0.0400 |
| GPT-5.6 Terra | $0.00858 | $0.0164 | $0.0242 | $0.0320 | $0.0476 |
| GPT-5.2 | $0.00978 | $0.0189 | $0.0280 | $0.0371 | $0.0553 |
| GPT-5.4 | $0.0107 | $0.0205 | $0.0302 | $0.0400 | $0.0595 |
| GPT-5.6 Sol | $0.0146 | $0.0276 | $0.0406 | $0.0536 | $0.0796 |
| GPT-5.5 | $0.0215 | $0.0410 | $0.0604 | $0.0799 | $0.1189 |
| GPT-6 Astra | $0.0364 | $0.0689 | $0.1014 | $0.1339 | $0.1989 |
| GPT-5.6 Cyber | $0.0536 | $0.1024 | $0.1511 | $0.1999 | $0.2974 |
| GPT-5.5 Cyber | $0.0536 | $0.1024 | $0.1511 | $0.1999 | $0.2974 |
| GPT-5 Pro | $0.0838 | $0.1618 | $0.2398 | $0.3179 | $0.4738 |
| GPT-5.2 Pro | $0.1174 | $0.2266 | $0.3358 | $0.4450 | $0.6634 |
| GPT-5.5 Pro | $0.1287 | $0.2457 | $0.3627 | $0.4797 | $0.7137 |
| GPT-5.4 Pro | $0.1287 | $0.2457 | $0.3627 | $0.4797 | $0.7137 |
Cheapest first. These are list rates applied to a token model — see the methodology for exactly where that diverges from an invoice.
The 1,000w column is the per-1,000-words figure people quote: $0.00054 on GPT-5 nano at these settings.
Providers do not price the same way, and the differences change which model wins.
OpenAI has by far the widest range of any provider we track — the most expensive model costs hundreds of times the cheapest per article. Sol, Terra and Luna have separate prices. Reasoning settings and generated token counts can still change a bill; this word-based estimate does not add a separate reasoning allowance. The newest entry is GPT-6 Astra, released 2026-09-03 and priced level with Anthropic's top model rather than against GPT-5.5.
Fast pricing is model-specific (roughly 1.67×–2.5× the Standard rate across the tracked catalog). Flex uses the half-price rate with slower synchronous scheduling; Batch is asynchronous at 50% off. Availability is model-specific, not a universal Pro-family rule: check the calculator and model documentation. These estimates exclude regional pricing uplifts; Astra Fast is unavailable with EU data residency.
The 1,050K-context models carry a threshold most calculators miss: a prompt over 272K input tokens bills at 2× input and 1.5× output for the whole session, not just the excess. Article workflows sit far below it, but a retrieval pipeline that stuffs a research corpus into context can cross it without anyone noticing.
Two things. Three Pro-tier models publish no cache-read rate at all, so prompt caching cannot reduce their input cost — if a workflow depends on caching a large shared system prompt, that rules them out regardless of quality. And GPT-5.6 Sol's current $4 / $20 is promotional, published as available only at least through 21 November 2026 with no successor rate named, so an annual budget built on it is provisional.
Every gallery below includes an unedited GPT-5 nano and GPT-5.5 draft, published blind with a current-rate cost comparison. The model and the price stay hidden until you reveal them, so this page can name who took part without spoiling which draft is which.
USD per million tokens, cheapest output first.
| Model | Input | Cached in | Output | Context |
|---|---|---|---|---|
| GPT-4.1 nano | $0.10 | $0.025 | $0.40 | 1,048K |
| GPT-5 nano | $0.05 | $0.005 | $0.40 | 400K |
| GPT-6 Luna Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $0.10 | $0.010 | $0.50 | 1,050K |
| GPT-4o mini | $0.15 | $0.075 | $0.60 | 128K |
| GPT-5.6 Luna Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $0.20 | $0.020 | $1.20 | 1,050K |
| GPT-5.4 nano | $0.20 | $0.020 | $1.25 | 400K |
| GPT-4.1 mini | $0.40 | $0.100 | $1.60 | 1,048K |
| GPT-5 mini | $0.25 | $0.025 | $2.00 | 400K |
| GPT-5.4 mini | $0.75 | $0.075 | $4.50 | 400K |
| GPT-4.1 | $2.00 | $0.500 | $8.00 | 1,048K |
| GPT-6 Sol Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $2.00 | $0.200 | $10.00 | 1,050K |
| GPT-5.1 | $1.25 | $0.125 | $10.00 | 400K |
| GPT-5 | $1.25 | $0.125 | $10.00 | 400K |
| GPT-4o | $2.50 | $1.250 | $10.00 | 128K |
| GPT-5.6 Terra Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $2.00 | $0.200 | $12.00 | 1,050K |
| GPT-5.2 | $1.75 | $0.175 | $14.00 | 400K |
| GPT-5.4 Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $2.50 | $0.250 | $15.00 | 1,050K |
| GPT-5.6 Sol Promotional rate, published as available at least through 2026-11-21; OpenAI has not announced the rate that follows it. Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $4.00 | $0.400 | $20.00 | 1,050K |
| GPT-5.5 Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $5.00 | $0.500 | $30.00 | 1,050K |
| GPT-6 Astra Prompts over 272K input tokens bill at 2x input / 1.5x output for the whole session. | $10.00 | $1.000 | $50.00 | 1,050K |
| GPT-5.6 Cyber Security-focused tier, gated behind the Daybreak program. | $12.50 | $1.250 | $75.00 | 1,050K |
| GPT-5.5 Cyber Security-focused tier, gated behind the Daybreak program. | $12.50 | $1.250 | $75.00 | 1,050K |
| GPT-5 Pro | $15.00 | — | $120.00 | 400K |
| GPT-5.2 Pro | $21.00 | — | $168.00 | 400K |
| GPT-5.5 Pro | $30.00 | — | $180.00 | 400K |
| GPT-5.4 Pro | $30.00 | — | $180.00 | 1,050K |
Source: OpenAI's published pricing. Compare against the other providers on the full pricing table.
Answers specific to this provider's rate structure.
GPT-5 nano is the cheapest tracked model from any provider for a full-length draft. It is priced for classification and extraction work rather than long-form prose, so treat its output as a first draft that needs real editing.
They are service levels for the same model, but not every model supports every tier. OpenAI Fast rates vary by model in the tracked catalog; Flex and Batch are 50% off where listed. Batch is asynchronous with up to 24-hour turnaround, while Flex is synchronous with slower scheduling. The calculator exposes only the tiers published for the selected model.
No. OpenAI replaced the effort dial with separate models: Sol, Terra and Luna are the effort tiers and each publishes its own rate. Changing an effort setting does not change per-token billing, though a model that emits more reasoning tokens will bill more output.
On the 1,050K-context models, a prompt over 272K input tokens bills at 2x input and 1.5x output for the entire session — not just for the tokens above the line. Article workflows sit far below it, but a retrieval pipeline that loads a large corpus can cross it unnoticed.
The Pro-tier models publish no cache-read rate, so prompt caching cannot reduce their input cost. We leave the column empty rather than assuming a percentage. If your workflow depends on caching a large shared prompt, those models are ruled out on structure regardless of quality.
Price your workflow here, then check it against the other providers before committing. At article volumes the difference between two reasonable models is usually smaller than a single hour of editing. When the work can run asynchronously, compare OpenAI's batch rate — not its standard rate — against DeepSeek, where no Batch tier is published. DeepSeek instead offers a separate off-peak discount.