What is xAI Grok 4.5?
Grok 4.5 is the model SpaceXAI documents as its intelligent coding model for agentic software, engineering and workflow tasks, trained in the company’s own data centres in Memphis on new datasets spanning science, engineering and mathematics. Two labels point at the same model and both are correct in their place: the developer’s pages write the company name SpaceXAI and the model name with no hyphen — Grok 4.5 — while this site’s own label, the one the navigation and the heading show, reads xAI Grok 4.5. The identifier settles which model is meant, and it accepts grok-4.5; two aliases resolve to it as well, grok-4.5-latest and grok-build-latest, the second of which is the name the document product calls it. The model page documents a context window of 500,000 tokens, text and image as inputs with text as output, function calling, structured outputs, reasoning switched on with no documented way to disable it, four supported reasoning efforts — low, medium, high and extra-high — with a default of high, and Batch API as not supported. Its rate limits sit on that same page, at 150 requests per second and 50,000,000 tokens per minute, and the documented regions are us-east-1 and us-west-2. The release notes date the launch to 16 July 2026 and quote the same two headline rates the announcement does. Three later facts belong in this paragraph rather than in a footnote. Grok 4.6 shipped on 12 August 2026 and Grok 4.7 on 21 September 2026, so this is no longer the newest model in its family even though it stays listed and priced in the official model table. The current selection guidance routes code and general work to Grok 4.7, which has a dedicated overview page that this model does not. And the deprecation list names Grok 4.5 nowhere: its two entries are a voice-transcription model reaching end of life on 2 October 2026 and an image-quality model retiring on 2 November 2026. One sentence needs its date attached before it is quoted. The launch announcement calls Grok 4.5 the company’s strongest model ever, and that was written on 16 July 2026; the model page now recommends a later release, so the phrase records a moment rather than a standing rank.
What xAI Grok 4.5 is worth using for
Two price tiers and one rule that reprices the whole request rather than the excess
The model page prints six figures and one rule, and the rule is the part that changes a bill. At or below 200,000 prompt tokens the standard rate is $2.00 per million input tokens, $0.30 per million cached input tokens and $6.00 per million output tokens. From 200,000 prompt tokens upward the same three figures become $4.00, $0.60 and $12.00, and the official sentence states that a request reaching that line is billed at the higher rate for all tokens in the request, not only for the ones past it. The launch announcement quotes the two headline numbers and stops there. Of the twenty-three competitor pages this research could read, sixteen printed the base input and output rates, fourteen mentioned cached input, and only two mentioned the long-context rule at all, which is the gap this section fills. Two readings the table invites are worth correcting. Because input, output and cached input are priced separately, a workload that resends a large stable prefix is billed differently from one that sends fresh text every time, and cached input at $0.30 is the cheapest line on the page by a factor of ten. And nothing on the model page attaches an end date to any of these rates; the single dated price sentence in the surrounding documentation belongs to a different member of the family.
A model is documented at $2 per million input tokens and $6 per million output tokens below 200,000 prompt tokens, and at $4 / $0.60 / $12 above that line, where the higher rate applies to every token in the request. Explain how a 240,000-token request is billed compared with a 190,000-token one, say what changes once the long prefix is cached, and state which of the two requests a prompt-caching strategy actually helps.
Four reasoning levels are documented, one of them is downgraded in writing, and this box has no level control
Three official pages describe reasoning for this model and they do not say the same thing, so all three are kept here. The model page lists four supported reasoning efforts — low, medium, high and extra-high — with high as the default. The release notes for the launch describe configurable reasoning effort as low, medium or high and do not mention extra-high. The capabilities guide names the fourth value and then qualifies it: extra-high is available on Grok 4.6 and later, and on models that do not support it, such as Grok 4.5, a request carrying extra-high is treated as high. Read together, those are two different registers of fact rather than a contradiction to be resolved — which values the API accepts, and what happens to one of them on this model. The same capabilities guide states that reasoning cannot be disabled. On this site the answer runs the other way: the live catalog marks reasoning as unsupported for Grok 4.5, so the composer’s thinking control is disabled rather than merely switched off, and no effort value can be selected or sent. Three statements, then, and this page keeps all three — the model page’s four values, the documentation’s downgrade rule, and this workspace’s absence of any reasoning control.
Three official sources describe one model differently. The model page lists four reasoning effort values including extra-high. The capabilities guide says extra-high is available on a later model and that, on this one, a request asking for extra-high is treated as high. The launch release notes list three values and no extra-high. Write the paragraph that presents all three without choosing one, then state separately what a request asking for extra-high actually receives.
The efficiency claim arrives with the number that backs it, and with a note on whose measurement it is
The announcement makes three claims about how much output the model needs. It is served at fast-model speeds of 80 tokens per second; it reaches twice greater token efficiency than the latest leading models at the same tasks; it solves tasks in under half the number of steps. It then gives one figure that can be checked: on SWE Bench Pro, Grok 4.5 averages 15,954 output tokens against 67,020 for the model the page names as Opus 4.8 (max), which it describes as 4.2 times fewer. Three caveats travel with that. The same page carries a footnote stating that competitor figures are drawn from the respective developers’ published system cards or benchmark leaderboards, so the comparison is assembled by the vendor rather than run head to head. The throughput figure describes how the model is served, not how this page behaves — this workspace publishes no throughput guarantee, and folding 80 tokens per second into a promise about response times here would add something neither page says. And the announcement’s own benchmark table prints rows where this model sits behind a competitor, so the efficiency summary and the full table are both part of the same document.
A launch post claims a model is served at 80 tokens per second, is twice as token-efficient as leading models at the same tasks, and solves tasks in under half the number of steps, then reports that on one benchmark it averages 15,954 output tokens against 67,020 for a named competitor on the same benchmark. Separate in writing what is measured, what is measured by the vendor from published cards rather than head to head, and what describes serving rather than the model itself.
What the official pages document for this model, and what this box actually opens
Two sets of statements describe this model and they answer different questions, so they are listed apart. On the official side the model page documents a 500,000-token context window; text and image as inputs with text as output; function calling and structured outputs; reasoning enabled with no documented off switch and four effort values; Batch API not supported; 150 requests per second and 50,000,000 tokens per minute; and the regions us-east-1 and us-west-2. It states no maximum output at all for this model, and the absence is not a very large number — a sibling model’s page carries an explicit no-limit line and this one does not. On this site’s side the page behind this chat box opens text chat with streaming replies, a web-search switch billed per search, and image and PDF attachments up to 24 MB each, with a reference price published in points. The gaps are the useful part. No context figure appears anywhere in the interface, because the catalog entry publishes no context length, so the official 500,000-token window describes the API rather than this page. No reasoning control exists at all, because the catalog marks reasoning unsupported for this model, so none of the four documented levels can be chosen here. And every reply is requested with at most 4,096 output tokens, because the catalog declares no maximum output and the composer falls back to 4,096 and caps whatever it reads at the same number. Note the direction of the reasoning gap: the model page says reasoning is enabled and cannot be disabled, while this site’s catalog says reasoning is unsupported. Those are claims about two different things — the model’s capability and this box’s controls — and only the second governs what you can press.
Write a product-page section that keeps two vocabularies apart: what a vendor’s API documentation states for a model (context window, input and output modalities, reasoning levels, tool support, rate limits) and what a hosted chat box actually exposes (a text box, an attachment control, a search toggle). Name each gap explicitly, and include two ceilings the hosted service sets for itself rather than inheriting from the model.
Official xAI Grok 4.5 specifications
- Developer
- SpaceXAI
- Model ID
- grok-4.5
- Official context window
- 500,000 tokens
- Official maximum output
- Not stated on the model page
- Official inputs
- Text and image
- Output
- Text
xAI Grok 4.5 and xAI Grok 4.6 side by side
Every value below comes from a single page — the Grok 4.6 announcement on x.ai — and from a single table within it. That table prints a Grok 4.6 High column and a Grok 4.5 High column across ten evaluations, so one official page proves both columns of every row, and all ten are reproduced here in the order the official table prints them. Read the direction before the size of the gaps: the left column is this page’s model and the middle column is its successor, and the successor scores higher on all ten rows. That is unusual for a landing page and it is deliberate — a table that kept only favourable rows would not be the official table. Two further columns exist on the same page, for GPT-5.6 Sol Max and Fable 5 Max, and are not reproduced. Two facts travel with the numbers rather than being dropped. The page’s own footnote states that the best score per evaluation is shown in bold and that third-party model scores are the best of self-reported or publicly available results, so not every figure in that table was measured by the vendor. And one value in this table disagrees with a value published elsewhere — the DeepSWE v1.1 figure for this model is 54% here and 53% in the launch announcement, which attributes its number to a different harness. Both are official, they are not reconciled, and the table value is the one shown. Every figure is a vendor evaluation of the models, not a measurement of this website, and no row predicts the quality or speed of the conversation you are about to have.
| Official evaluation | xAI Grok 4.5 | xAI Grok 4.6 |
|---|---|---|
| AA Intelligence Index | 56 | 61 |
| GDPVal-AA v2 | 1526 | 1753 |
| CursorBench v3.2 | 66.7% | 69.9% |
| DeepSWE v1.1 | 54% | 65.9% |
| FrontierCode v1.1 (Extended) | 56.6% | 61.3% |
| APEX-Agents | 47.1% | 57.5% |
| Terminal-Bench v3.0 | 15.7% | 26% |
| APEX-SWE | 53.6% | 56.4% |
| AA-Briefcase | 1313 | 1577 |
| Harvey LAB (Vals) | 12.9% | 15.8% |
How to use xAI Grok 4.5 online
Choose the model
This page opens with xAI Grok 4.5 selected. The model navigation on the left and the selector inside the chat box stay in sync, switching models keeps the current conversation, and starting a new chat is how you get a clean context. Three naming facts are worth fixing before you send anything. Two labels point at this model and both are correct in their place: the developer’s pages write the company name SpaceXAI and the model name without a hyphen, and this site’s display label, which is what the navigation and the heading show, reads xAI Grok 4.5. The identifier the API accepts is grok-4.5, and the two aliases grok-4.5-latest and grok-build-latest resolve to the same model. And the official model page carries no snapshot list, so there is no dated variant of this identifier to fall back to.
Decide what you expect from reasoning, because this box has no reasoning control
The official pages document four reasoning efforts for this model — low, medium, high and extra-high — with high as the default, and one of them is qualified: the capabilities guide states that extra-high is available on Grok 4.6 and later and that, on a model such as Grok 4.5, a request carrying it is treated as high. The same guide states that reasoning cannot be disabled. What this page gives you is neither of those. The catalog entry marks reasoning as unsupported for this model, so the thinking control renders disabled and no effort value can be chosen or sent. There is therefore no lever here for reasoning at all, and the only way to shape how much thinking a task gets is the prompt itself. Note that this is not the same statement as the two official ones: the model can reason, the API accepts four values, and this box cannot ask for any of them.
Sign in, send, and check the cost against the official pages
Sending a message needs an account with available points. The composer reads a reference price from the live model catalog and a finished request reports what it actually cost. One ceiling is set by this site rather than by the model, so plan around it: replies are requested with at most 4,096 output tokens, and the catalog entry declares no maximum output, so the composer supplies the limit. Because the rates quoted on this page are the rates the official pages print, re-check them there before relying on a figure for planning — and treat any third-party summary that presents a relay, resale or promotional figure as out of scope, because no official page publishes one. Nothing here converts points into dollars, and nothing on the official pages sets an expiry on the rates.
What xAI Grok 4.5 costs, and why this page carries no end date
The first thing to establish about this model’s price is what is not attached to it. No official page puts an end date on any of its rates, and the one dated price sentence in the surrounding documentation belongs to a different member of the family. Both the model page and the announcement were checked on 3 October 2026. At or below 200,000 prompt tokens the standard rate is $2.00 per million input tokens, $0.30 per million cached input tokens and $6.00 per million output tokens. From 200,000 prompt tokens upward the same three figures become $4.00, $0.60 and $12.00, per million tokens each. The stated rule is that a request whose prompt reaches 200,000 tokens is billed at the higher rate for all tokens in the request, so crossing the line reprices the entire call rather than only the part above it. Two details follow from the way the table is laid out. Cached input is priced at $0.30 per million tokens below the line, which is a tenth of the uncached input rate. And Batch processing is not a discount tier here at all: the model page records Batch API as not supported, and the platform pricing table applies its batch discount to four other models with a sentence stating that models not listed have no batch discount. The other charges on that page are for platform features rather than for this model: web search at $5 per thousand calls, X search at $5 per thousand posts fetched and $10 per thousand user profiles fetched, code execution at $5 per thousand calls, file attachments at $5 per thousand calls, collections search at $2.50 per thousand calls, file storage at $0.025 per GiB per day, collections storage at $0.10 per GiB per day, priority processing at twice the standard token rate, and the US regional endpoint at 1.1 times the token rate for the two models it serves — a list this model is not on. Where this model sits in its own generation is arithmetic rather than interpretation. The official model table puts 4.5 and 4.6 at identical input and output rates and separates them only on cached input, where this model is cheaper. So the price table does not encode the capability gap the comparison table above measures: 4.6 outscores 4.5 on all ten official rows and costs the same per input and output token, while 4.5 is the cheaper of the two when a large prefix is cached. None of the figures above is this site’s checkout price. The workspace bills points, and the catalog publishes a reference price of 229 points per million input tokens, 687 points per million output tokens and 34.5 points per million cached read tokens, with no conversion between that basis and the dollar rates.
View points and plans ↗xAI Grok 4.5 FAQ
What does xAI Grok 4.5 cost, and will the price change?
The rates the official pages print are $2.00 per million input tokens, $0.30 per million cached input tokens and $6.00 per million output tokens at or below 200,000 prompt tokens, and $4.00, $0.60 and $12.00 above that line, where the higher rate applies to all tokens in the request. On whether it will change: this page attaches no end date, because the official sources give none — the model page prints rates and rules only, and the one dated price sentence in the surrounding documentation names a different model. Batch processing is not available for this model at all, so there is no batch discount to quote. None of these figures is this site’s checkout price: the workspace bills points, at 229 points per million input tokens, 687 points per million output tokens and 34.5 points per million cached read tokens, with no conversion between the two bases.
How does Grok 4.5 compare with Grok 4.6 on SpaceXAI’s own table?
Across the ten evaluations that one official page prints for both models, Grok 4.6 scores higher on all ten. The gaps run from 3.2 points on CursorBench v3.2 to 10.3 points on Terminal-Bench v3.0, with the largest relative separation on APEX-Agents. That table’s footnote notes that the best score per evaluation is shown in bold and that third-party model scores are the best of self-reported or publicly available results, so not every figure was measured by the vendor itself. One value disagrees with a value published elsewhere: the DeepSWE v1.1 figure for this model is 54% in that table and 53% in the launch announcement, which attributes its number to a different harness. Both are official and neither is treated here as the correction of the other. All of these are vendor evaluations of the models, not measurements of this website.
What do the official pages document for this model?
The model page records a 500,000-token context window, text and image inputs, text output, function calling, structured outputs, reasoning enabled with no documented way to disable it, four reasoning efforts (low, medium, high and extra-high) with a default of high, Batch API not supported, rate limits of 150 requests per second and 50,000,000 tokens per minute, and the regions us-east-1 and us-west-2. Three of those are easy to misread. The model page states no maximum output at all for this model, which is an absence rather than a large number — a sibling model’s page carries an explicit no-limit line and this one does not. The regions field describes where inference lands, not the optional US regional endpoint, which serves two other models. And no knowledge cutoff is published for this model, so any cutoff figure quoted for it is not from the official pages. This page runs against none of those numbers directly: replies here are requested with at most 4,096 output tokens, and no context figure is displayed.
Which reasoning levels exist, and which one runs here?
Three official pages describe this and they differ, so all three are kept. The model page lists four supported reasoning efforts — low, medium, high and extra-high — with high as the default. The launch release notes describe configurable reasoning effort as low, medium or high and omit extra-high. The capabilities guide names extra-high and then qualifies it: it is available on Grok 4.6 and later, and on models that do not support it, such as Grok 4.5, a request carrying extra-high is treated as high. The same guide states that reasoning cannot be disabled. What runs here is neither of those: this site’s catalog marks reasoning as unsupported for this model, so the thinking control is disabled and no effort value can be selected or sent at all.
Is Grok 4.5 still the newest model in its family?
No, and the official pages carry the dates either way. Grok 4.6 shipped on 12 August 2026 and Grok 4.7 on 21 September 2026, while Grok 4.5 dates to 16 July 2026. The current selection guidance routes code and general work to Grok 4.7, which has a dedicated overview page that this model does not. Two things point the other way and both are also official. Grok 4.5 remains listed and priced in the model table, and the deprecation list does not mention it anywhere — its two entries are a voice-transcription model reaching end of life on 2 October 2026 and an image-quality model retiring on 2 November 2026. And the launch announcement’s sentence calling this the company’s strongest model ever is dated 16 July 2026, so read today it records a moment rather than a ranking.
Is Grok 4.5 free, and are relay or resale prices official?
Neither, in the sense the question usually means. The launch announcement does contain a sentence about free Grok 4.5 usage for a limited time in its document product and in one editor, but the sentence carries no date range anywhere on that page, and this site does not publish eligibility claims whose window cannot be checked — so no free-access claim appears on this page. On relay and resale prices: no official page publishes one. The rates quoted in this article are the ones printed by the developer’s own model and pricing pages, and everything else — relay services, mirror sites, unregistered facsimiles — is outside the official record. Where this site is concerned, billing is in points against the catalog reference price, and there is no conversion between points and dollars.
What does this page expose compared with the official API?
Less than the official pages document, and the difference is listed rather than glossed over. Available here: text chat with streaming replies, image and PDF attachments up to 24 MB each, and a web-search switch billed per search. Not available here: any context figure in the interface, because the catalog entry publishes no context length; any reasoning control at all, because the catalog marks reasoning unsupported for this model, so none of the four documented effort values can be chosen; the two aliases, with no way to address the model by either; function calling and structured outputs; the endpoint type; the two regions; the rate limits, which are not exposed as any kind of quota display; and Batch processing, which the official model page does not support in any case. One ceiling is set by this site rather than by the model: replies are requested with at most 4,096 output tokens. Pricing is in points and is not a conversion of the official dollar rates.
Is NanoBanana Pro the official SpaceXAI website?
No. NanoBanana Pro is an independent platform providing a shared workspace for multiple AI models, and this page is a guide to one of them rather than a product page from its developer. SpaceXAI develops and publishes Grok 4.5 and its successors; the official model page, the launch announcement, the model and pricing table, the reasoning capabilities guide, the platform pricing page, the release notes, the regional endpoint page and the Grok 4.6 announcement are all linked at the bottom of this page, and every specification, price and evaluation figure quoted here comes from one of them. If you need the developer’s own contract, rate limits or support, follow those links rather than relying on this page.
Sources and model information
Information checked
Start an xAI Grok 4.5 conversation
The chat box above runs this model, and the guide on this page is written to be checked rather than trusted: the specifications come from the official model page, the comparison comes from the single official page that prints both columns, and the places where official pages disagree with each other or with this workspace are named rather than smoothed over. Send the task the way you would brief a person — outcome, constraints, evidence, stopping condition — and re-check the figures against the pages linked below, since this model’s rates carry no end date and its successor has already been priced at the same headline input and output rates.
Start a chat with xAI Grok 4.5 ↑