At the frontier tier, OpenAI and Anthropic now charge exactly the same: GPT-6 Astra and Claude Fable 5.1 are both $10 per million input tokens and $50 per million output. The mid tier matches too — GPT-6.1 Sol and Claude Sonnet 5.5 are both $2 and $10.
The budget tier matches too since 7 October: GPT-6 Luna and Claude Haiku 5.5 both cost $0.10 input and $0.50 output — about ten times less than Claude Haiku 4.5. Haiku 5.5’s rate applies to prompts up to 100,000 tokens. Use the calculator below to apply these rates to your own request volume, token sizes and cache hit rate.
Enter your own volumes. Costs update as you type, and are ranked cheapest first.
| Model | Provider | Cost / month | Per 1k requests |
|---|
Published rates per million tokens, verified 21 October 2026 against each provider’s own pricing page. Standard tier, short context.
| Model | Provider | Input | Cached in | Cache write | Output |
|---|---|---|---|---|---|
| GPT-6 Astra | OpenAI | $10.00 | $1.00 | $12.500 | $50.00 |
| Claude Fable 5.1 | Anthropic | $10.00 | $0.25 | $12.500 | $50.00 |
| GPT-5.6 Cyber | OpenAI | $12.50 | $1.25 | $15.625 | $75.00 |
| Claude Opus 5.5 | Anthropic | $4.00 | $0.20 | $5.000 | $20.00 |
| GPT-5.6 Sol | OpenAI | $4.00 | $0.40 | $5.000 | $20.00 |
| GPT-6.1 Sol | OpenAI | $2.00 | $0.10 | $2.500 | $10.00 |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $0.20 | $2.500 | $10.00 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $0.10 | $1.250 | $5.00 |
| Claude Haiku 5.5 | Anthropic | $0.10 | $0.01 | $0.125 | $0.50 |
| GPT-6 Luna | OpenAI | $0.10 | $0.01 | $0.125 | $0.50 |
The most striking thing in the table above is not any single price. It is that OpenAI and Anthropic have converged on identical numbers at every tier:
When two competitors price to the cent at matching tiers, price has stopped being a differentiator at those tiers. The decision moves entirely to output quality, reliability and latency on your specific workload — which you can only establish by running both.
Output runs about five times the input rate across every model in the table. If your bill is high and you have not looked at response length, that is almost always the cheapest fix available — and it is a one-line prompt change rather than a migration.
Cached input on GPT-6 Astra is $1.00 against $10.00. On Claude Fable 5.1 it is $0.25 against $10.00 — a 97.5% reduction on the repeated portion. Any workload with a stable prefix (a long system prompt, a document, a fixed tool list) should be using it.
The constraint is that caching is a prefix match. One changed byte early in the prompt invalidates everything after it, so a timestamp or per-request ID near the top of a system prompt silently destroys the discount. Put stable content first, volatile content last, and check that your cache-read token count is actually non-zero.
Both providers price batch processing at 50% of standard rates. The trade is latency, not quality. Anything that does not need an answer this second — overnight classification, bulk enrichment, backfills — is leaving half the money on the table if it runs synchronously.
The calculator above compares token costs, because that is what is knowable in advance. It is not the number that decides anything.
A model at five times the token price that succeeds on the first attempt is cheaper than one that needs four attempts and a human review. The reverse is just as often true on simple work. The only way to know which applies to you is to assemble thirty to fifty real examples from your own workload, run every candidate against them several times, and record quality, variance, latency and total spend together.
That takes an afternoon and it settles the question permanently — unlike any pricing table, including this one.
Price has stopped separating the two big labs at the frontier, mid and budget tiers — they are matched to the cent. Choose on quality and reliability for your workload, because that is the only axis still in play.
Where price still decides things is in the levers you control: caching, batch processing and response length will move your bill further than switching provider will.
Not every AI decision needs a spreadsheet. XAIWebsites generates your complete business website — copy, layout, mobile version and a downloadable site (managed hosting optional) — from a short questionnaire. $199 one-time, and you preview it before you pay.
OpenAI's published rate is $10.00 per million input tokens and $50.00 per million output tokens, with cached input at $1.00 and cache writes at $12.50. Claude Fable 5.1 sits at exactly the same $10 and $50, though its cached input is cheaper at $0.25.
Two models tie at $0.10 input and $0.50 output per million tokens: OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5, released on 7 October 2026. Both are roughly ten times cheaper than Claude Haiku 4.5 at $1 and $5. One difference matters: Anthropic's $0.10 / $0.50 rate applies to prompts up to 100,000 tokens; above that Haiku 5.5 bills at $0.50 input and $2.50 output.
Anthropic lists $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, with cache reads at $0.01 and 5-minute cache writes at $0.125. For prompts over 100,000 tokens the rates rise to $0.50 input and $2.50 output. Batch processing takes 50% off. Unlike other Claude 4.6-and-later models, Haiku 5.5 does not include the full 1M-token context window at one flat price.
They are identically priced: $2.00 per million input tokens and $10.00 per million output tokens. OpenAI's cached input is cheaper at $0.10 against $0.20. At that point price is not the deciding factor — run both against your own evaluation set and choose on output quality and reliability.
Output tokens are generated one at a time and cannot be batched the way input processing can, so they cost providers considerably more to serve. Across every provider output runs about five times the input rate, which is why shortening responses is usually the fastest way to cut an AI bill.
Substantially, for any workload with a stable prompt prefix. Cached input on GPT-6 Astra is $1.00 against $10.00, and on Claude Fable 5.1 it is $0.25 against $10.00 — a 90% to 97.5% reduction on the repeated portion. The catch is that caching is a prefix match, so a timestamp near the top of a system prompt silently destroys the discount.
It applies each provider's published per-token rates to the volumes you enter, so the arithmetic is exact. Real bills differ because of retries, tool calls, reasoning tokens on models that use them, long-context surcharges above certain prompt sizes, and regional tax or data-residency uplifts. Treat the result as a planning estimate, not a quote.
Both OpenAI and Anthropic price batch processing at 50% of standard rates. The trade is latency — batch jobs are processed asynchronously rather than in real time. For anything that does not need an immediate answer, such as overnight classification or bulk enrichment, it is the single easiest 50% saving available.
Rates verified 21 October 2026. Both providers change pricing without much notice, so confirm before committing to a budget.
Displayed provider pricing can include regional tax or data-residency uplifts depending on where you are billed. The figures above are the standard published list rates.
Rates verified 21 October 2026 against each provider’s own pricing pages. Model pricing changes frequently — confirm before budgeting. See our methodology.