The two most expensive models in this table cost exactly the same. OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 are both $10 per million input tokens and $50 per million output.OpenAI, Pricing · Anthropic, Pricing, both read at source 16 Sep 2026: gpt-6-astra “$10.00” input and “$50.00” output; Claude Fable 5.1 “$10 / MTok” and “$50 / MTok”. Below them the four vendors diverge by more than an order of magnitude. Here is the whole board, read off each company’s own pricing page on 16 September 2026.
One table, four vendors, no third-party trackers.
- Top of this table is $10 in, $50 out — and OpenAI and Anthropic price it identically.OpenAI, Pricing · Anthropic, Pricing, both read at source 16 Sep 2026:
gpt-6-astra“$10.00” input and “$50.00” output; Claude Fable 5.1 “$10 / MTok” and “$50 / MTok”. - OpenAI’s small models are the cheapest here — GPT-5 nano is $0.05 in and $0.40 out. Google’s cheapest in the table, Gemini 3.8 Flash, is $0.75 and $3.75 until 31 December.OpenAI, Pricing, read at source 17 Sep 2026:
gpt-5-nano“$0.05” input and “$0.40” output · Google, Gemini Developer API pricing, read 16 Sep 2026: Gemini 3.8 Flash input “$0.75 through December 31, 2026”, output “$3.75 through December 31, 2026”. - xAI sits in the middle — Grok 4.6 at $2.00 in and $6.00 out, Grok 4.3 at $1.25 and $2.50.xAI, Pricing, read at source 17 Sep 2026: grok-4.6 “$2.00” input and “$6.00” output, grok-4.3 “$1.25” and “$2.50”, below 200k prompt tokens. Grok 4.7, released 21 Sep 2026, is now xAI’s recommended model at the same price as 4.6 (see Grok 4.7 vs 4.6).xAI, Models, read at source 22 Sep 2026: “For everything else, including code, use Grok 4.7. It is the most capable model we’ve built.”
- Three of the four charge more for long prompts; Anthropic does not. xAI doubles input and output. OpenAI and Google double input and raise output by half, on the models where they publish a long-context rate.xAI, Pricing: grok-4.6 “$4.00” input and “$12.00” output at or above 200k prompt tokens · OpenAI, Pricing:
gpt-6-astralong context input “$20.00” and output “$75.00” · Google, Gemini Developer API pricing: Gemini 3.1 Pro output “$18.00, prompts > 200k” · Anthropic, Pricing: “Claude 4.6 and later models and Claude Mythos Preview include the full 1M token context window at standard pricing.” All read at source 17 Sep 2026. - All four discount cached input, but not by the same amount. A tenth of the input price at OpenAI and Google and on most Claude models; less off than that at xAI, where Grok 4.6 caches at a quarter.Anthropic, Pricing: “Cache read (hit)” at “0.1x base input price (0.025x on Claude Fable 5.1 and Claude Mythos 5.1)” · OpenAI, Pricing:
gpt-6-astracached input “$1.00” against “$10.00” · Google, Gemini Developer API pricing: Gemini 3.1 Pro context caching “$0.20, prompts <= 200k tokens” against input “$2.00, prompts <= 200k tokens” · xAI, Pricing: grok-4.6 cached input “$0.50” against “$2.00”. All read at source 17 Sep 2026.
Every price here was correct when this site last checked it, between 16 Sep and 23 Sep 2026. Prices change often and this site no longer updates them, so check the vendor’s own page before you rely on one: OpenAI pricing · Anthropic pricing · Google pricing · xAI pricing.
The board
| Model | Vendor | Input | Output |
|---|---|---|---|
| GPT-6 Astra | OpenAI | $10.00 | $50.00 |
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 |
| Claude Opus 5 | Anthropic | $5.00 | $25.00 |
| GPT-5.6 Sol | OpenAI | $4.00 | $20.00 |
| Gemini 3.1 Pro | $2.00 | $12.00 | |
| Claude Sonnet 5 | Anthropic | $2.00 | $10.00 |
| Grok 4.6 | xAI | $2.00 | $6.00 |
| Grok 4.3 | xAI | $1.25 | $2.50 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 |
| Gemini 3.8 Flash | $0.75 | $3.75 | |
| GPT-5.6 Luna | OpenAI | $0.20 | $1.20 |
| GPT-5 nano | OpenAI | $0.05 | $0.40 |
OpenAI, Pricing · Anthropic, Pricing · Google, Gemini Developer API pricing (read in a browser; the docs refuse automated fetches) · xAI, Pricing. All read 16 Sep 2026.
gpt-6-astra “$10.00” input; Claude Fable 5.1 “$10 / MTok”. The other rows are the table above, each read off the vendor’s own pricing page the same day.A caveat on the bottom row. The snapshot behind GPT-5 nano is scheduled to shut down on 11 December 2026, with GPT-5.6 Luna named as its replacement — so the cheapest price here has an end date.OpenAI, GPT-5 nano model page, read at source 22 Sep 2026: “Default snapshot: gpt-5-nano-2025-08-07” · OpenAI, Deprecations, read at source 22 Sep 2026: shutdown date “Dec 11, 2026” for gpt-5-nano-2025-08-07, recommended replacement gpt-5.6-luna.
A newer Anthropic model since this table was read. Claude Opus 5.5, released on 22 Sep 2026, costs $4.00 in and $20.00 out, with a 1M-token context window, and Anthropic now tells developers to start with it. Claude Opus 5, in the table above, stays on sale as a legacy model at $5.00 and $25.00. The two are compared on Claude Opus 5.5 vs Opus 5.Anthropic, Models overview, read at source 23 Sep 2026: Claude Opus 5.5 “$4 / input MTok, $20 / output MTok”; “If you’re unsure which model to use, start with Claude Opus 5.5 for most workloads.”; and “Legacy models (still available)”, a list that now includes Claude Opus 5. Anthropic, Claude Opus 5.5, same day: “Released September 22, 2026.” and “Context window: 1M tokens”.
Two hundred times separates the top of this table from the bottom on input. The interesting question is almost never which model is best — it is which rung of the ladder each part of your workload actually needs.
Where they agree, and where they do not
Long prompts cost more at three of the four, and not by the same amount. OpenAI publishes a short and a long column for every current model: input doubles and output rises by half, $50.00 to $75.00 on Astra. Its model pages put the line above 272,000 input tokens and, like xAI, charge the higher rate on the whole request: “Prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request.”OpenAI, GPT-6 Astra model page, read at source 22 Sep 2026. The GPT-5.6 Luna model page, same day: “Prompts with >272K input tokens are priced at 2x input and 1.5x output for the full request.” Google does the same on Pro above 200,000 tokens — $4.00 instead of $2.00 in, $18.00 instead of $12.00 out. xAI doubles both, applies the higher rate to the whole request once a prompt crosses its threshold, and says so plainly: “requests whose prompt reaches the listed token threshold are billed at the higher rate for all tokens in the request.”xAI, Pricing, read at source 17 Sep 2026: grok-4.6 is “$2.00” input and “$6.00” output below 200k prompt tokens, “$4.00” and “$12.00” at or above it · OpenAI, Pricing, read at source 17 Sep 2026: gpt-6-astra short context output “$50.00”, long context output “$75.00” · Google, Gemini Developer API pricing, read 17 Sep 2026: Gemini 3.1 Pro input “$4.00, prompts > 200k tokens”, output “$18.00, prompts > 200k”. Anthropic is the exception: “Claude 4.6 and later models and Claude Mythos Preview include the full 1M token context window at standard pricing.”Anthropic, Pricing, read at source 17 Sep 2026, section “Long context pricing”.
Cached input is a tenth of fresh input at OpenAI and Google, and on most Claude models. Anthropic states the multiplier outright — “A cache hit costs 10% of the standard input price” — and goes lower on its top model: “On Claude Fable 5.1 and Claude Mythos 5.1, a cache hit costs 2.5% of the standard input price”.Anthropic, Pricing, read at source 17 Sep 2026: “Cache read (hit)” is “0.1x base input price (0.025x on Claude Fable 5.1 and Claude Mythos 5.1)”. Since 22 Sep 2026 Claude Opus 5.5 is a second exception, at a twentieth.Anthropic, Pricing, read at source 23 Sep 2026: “On Claude Opus 5.5, a cache hit costs 5% of the standard input price ($0.20 USD per million tokens).” OpenAI’s published figures work out to a tenth — $1.00 against $10.00 on Astra. xAI discounts less: $0.50 against $2.00 on Grok 4.6.OpenAI, Pricing: gpt-6-astra cached input “$1.00” against “$10.00” · xAI, Pricing: grok-4.6 cached input “$0.50” against input “$2.00”. Both read at source 17 Sep 2026.
Batch work is about half price. Anthropic offers “a 50% discount on both input and output tokens” on its Batch API, and OpenAI’s batch table is half its standard table on every current model.Anthropic, Pricing, read at source 16 Sep 2026: the Batch API allows “asynchronous processing of large volumes of requests with a 50% discount on both input and output tokens.”OpenAI, GPT-6 Astra model page, read at source 22 Sep 2026: “Batch and Flex are priced at 50% of Standard rates. Fast mode is priced at 2x the applicable rates.”First-hand: until 22 Sep 2026 the OpenAI half was labelled reasoning, from comparing its batch and standard pricing tables, because the pricing page does not state the ratio. The Astra model page does.
What changes soon
Two dated movements are already published rather than predicted:
- Google’s cheapest model in this table doubles on 1 January 2027 — Gemini 3.8 Flash goes to $1.50 and $7.50. Gemini 3.7 Flash, not in the table, carries the same price and the same increase. Both are printed on the pricing page as of 22 Sep 2026. Google’s cheaper Gemini 3.5 Flash-Lite, also not in the table, is listed at $0.30 in and $2.50 out.Google, Gemini Developer API pricing, read in a browser 22 Sep 2026: Gemini 3.8 Flash and Gemini 3.7 Flash each input “$0.75 through December 31, 2026. $1.50 starting January 1, 2027.” and output “$3.75 through December 31, 2026. $7.50 starting January 1, 2027.”; Gemini 3.5 Flash-Lite input “$0.30 (text / image / video / audio)”, output “$2.50”.First-hand: until 22 Sep 2026 this line said “Google’s cheapest model doubles on 1 January 2027”. Gemini 3.8 Flash is the cheapest Google model in this table, not Google’s cheapest; 3.5 Flash-Lite costs less.
- Anthropic cancelled a rise. Sonnet 5’s introductory $2/$10 became the standard price instead of climbing to $3/$15.Anthropic, Pricing, read at source 16 Sep 2026: the pricing “announced at launch as introductory pricing through August 31, 2026, is now the standard price.”
The detail on each is on Gemini vs Grok prices and which Claude model to use.
The number that is not on this page
Price per token is not price per job. A model that costs four times more and solves the task in one attempt is cheaper than one that costs a quarter and needs five. None of these vendors publishes tokens-per-task for anything, and this site has not measured it.
Use this table to rule options out on budget, then test the two or three that survive on your own work. The deeper breakdowns are on what the OpenAI API costs and the Claude lineup.
Free or paid? These are API prices; for the consumer chat plans see free vs paid AI chat.
What this page could not verify
- Meta’s token prices. Not published on its developer site, so Meta is absent rather than estimated.
- Any quality claim. This is a price table and nothing else.
- Enterprise and volume rates. All four route those to sales.
- Whether every OpenAI model uses the same long-context threshold. Google and xAI publish 200,000 on their pricing pages; OpenAI’s 272,000 was read on the Astra and Luna model pages only.Google, Gemini Developer API pricing, read in a browser 16 Sep 2026: “$4.00, prompts > 200k tokens” · xAI, Pricing, read at source the same day · OpenAI, GPT-6 Astra model page, read at source 22 Sep 2026: “more than 272K input tokens”.First-hand: until 22 Sep 2026 this item said OpenAI labels the columns without stating the boundary. True of the pricing page; the model pages state it.
OpenAI, Pricing · Anthropic, Pricing · Google, Gemini Developer API pricing · xAI, Pricing. All read at source on 16 September 2026.
No third-party price tracker was used for any figure here. API prices are the most volatile numbers on this site — re-read before you budget.