Grok 4.5 is SpaceXAI's frontier model for coding, agentic tasks, and knowledge work — and it shipped everywhere at once: grok.com, X, the iOS and Android apps, Cursor, the xAI API, and Microsoft Office add-ins for Word, Excel, PowerPoint, and Outlook. Priced at $2 per million input tokens and $6 per million output tokens with a 500K context window, it is the cheapest frontier-tier model on the market by a wide margin, and the first non-Microsoft AI to ship as the default inside the Office suite.
For most knowledge workers and developers, the practical takeaway is not the benchmark scores — it is that you no longer need to choose where to use Grok. The same model that writes your Word documents also runs in Cursor and answers on X, which means whatever prompting patterns you learn in one surface carry over to all the others. The catch: Office add-ins require a paid SuperGrok or Business plan, and the API is pay-per-token with extra costs for server-side tools like web search.
Last verified: 2026-07-30 · Best for coding agents and knowledge work · Cheapest frontier API at $2/$6 per 1M tokens · 500K context window · Available in Office, Cursor, and API
What is Grok 4.5 and why does the "everywhere" rollout matter?
Grok 4.5 is a large language model from SpaceXAI (formerly xAI), launched July 8, 2026, and trained on tens of thousands of NVIDIA GB300 GPUs in Memphis. It has a knowledge cutoff of February 1, 2026, a 500,000-token context window, and configurable reasoning effort (low, medium, or high, defaulting to high) (xAI docs).
The "everywhere" rollout is the real story. SpaceXAI pushed Grok 4.5 to grok.com, X, the iOS and Android apps, Cursor (all plans), the xAI API console, and the Microsoft Office add-ins — all on the same day. That means one model runs identically whether you are drafting in Word, debugging in Cursor, or asking a quick question on X from your phone. Whatever the model learns to do better, it does better everywhere on the same day, with no update to install and no version to track.
This matters because most AI tools still live in silos. You learn one set of prompting habits in ChatGPT, another in Copilot, and a third in your code editor. When the same model is running underneath all your surfaces, the prompting skills you build in one place compound across the rest.
How much does Grok 4.5 cost?
Grok 4.5's API pricing is $2.00 per million input tokens and $6.00 per million output tokens, with cached input at $0.30 per million tokens — about 85% cheaper than the standard input rate if your prompts repeat (xAI pricing docs). For requests exceeding 200K tokens in context, pricing doubles to $4 in / $12 out per million.
| Plan | Price | What you get |
|---|---|---|
| API (pay-per-token) | $2 in / $6 out per 1M tokens | Full model access, 500K context, function calling, web search, X search, code execution |
| Cached input | $0.30 per 1M tokens | 85% discount on repeated prompt prefixes |
| SuperGrok | $30/month | Chat interface access, ~100 prompts per 2-hour window |
| SuperGrok Heavy | $300/month | Extended reasoning, heavy agent use |
| Grok Business | $30/user/month | Team collaboration, centralized billing |
| X Premium+ | $8/month → includes Grok | Basic Grok access inside X |
The hidden cost layer is server-side tools. Web Search and X Search each cost $5 per 1,000 calls on the API. Code Execution is $5 per 1,000 calls. File attachment search is $10 per 1,000 calls. A search-heavy agent workflow can spend more on tool calls than on tokens (xAI pricing docs).
Compared to competitors at the API level: Claude Opus 4.8 charges $5 in / $25 out per 1M tokens, and GPT-5.5 charges $5 in / $30 out. Grok 4.5 is roughly 60% cheaper on input and 76% cheaper on output than the nearest frontier alternatives — though xAI's benchmark claims compare against Opus 4.7, not the current Opus 4.8.
How does Grok 4.5 perform on benchmarks?
Grok 4.5 scores competitively but is not the benchmark king. Here are the numbers from SpaceXAI's own announcement, with competitor figures drawn from their respective published system cards:
| Benchmark | Grok 4.5 | Top competitor | Source |
|---|---|---|---|
| SWE Marathon (pass@1) | 29.0% (1st place) | Opus 4.8 (max): 26.0% | xAI announcement |
| SWE Bench Pro | 64.7% | Fable (max): 80.4% | xAI announcement |
| Terminal Bench 2.1 | 83.3% | Fable (max): 84.3% | xAI announcement |
| DeepSWE 1.0 | 62.0% | Fable (max): 66.1% | xAI announcement |
The edge that SpaceXAI highlights is not raw scores but token efficiency. On SWE Bench Pro, Grok 4.5 resolves tasks using an average of 15,954 output tokens — approximately 4.2x fewer than Opus 4.8 (max), which uses 67,020 tokens on the same tasks (xAI announcement). Combined with 80 tokens-per-second serving speed, that means faster, cheaper completions per task even when the accuracy gap is modest.
Independent benchmark aggregators place Grok 4.5 at #4 on the Artificial Analysis Intelligence Index — behind Claude Opus 4.8 and GPT-5.5 on overall intelligence, but ahead on price-to-performance ratio.
What can Grok 4.5 do inside Microsoft Office?
Grok 4.5 ships as free add-ins for Microsoft Word, Excel, PowerPoint, and Outlook, available through the Microsoft AppSource marketplace. Each add-in appears as a conversation sidebar inside the app — you chat with it, hand it tasks, or ask it to analyze whatever document you have open. The add-ins require a paid SuperGrok, Heavy, Business, or Enterprise plan, and usage limits apply (Microsoft Marketplace: Grok for Word).
Here is what each add-in actually does:
| App | What Grok 4.5 does inside it |
|---|---|
| Word | Drafts structured prose, rewrites paragraphs, generates outlines, creates tables, expands ideas, summarizes long reports — all without leaving the document |
| Excel | Builds multi-sheet models with working formulas, pulls live web research into the spreadsheet, leaves sticky notes explaining its reasoning |
| PowerPoint | Builds slides using native shapes (not pasted images), creates complex diagrams, designs intuitive slide content — shapes stay editable after generation |
| Outlook | Triages inbox, drafts responses, summarizes email threads |
The key differentiator versus Microsoft's own Copilot is Grok's real-time data access. Grok can pull current news and trending topics through its X integration — something Copilot does not offer natively. In a demo shared by xAI, senior engineer Matthew Dabit turned a dense neuroscience research paper into a nine-slide, presentation-ready deck in under five minutes using native shapes (xAI announcement).
Microsoft 365 has over 400 million paid seats globally. Grok 4.5 is the first AI model from outside the Microsoft portfolio to ship as the default inside the Office suite, which puts xAI in direct competition with Microsoft's own Copilot inside Microsoft's own apps.
How do you use Grok 4.5 in Cursor and Grok Build?
Grok 4.5 is the default model in Grok Build, SpaceXAI's terminal-based coding agent and CLI. You install it with a single command:
curl -fsSL https://x.ai/cli/install.sh | bash
Grok Build supports plan mode (propose changes before writing code), subagents, MCP servers, AGENTS.md conventions, persistent memory, and git integration — features that will be familiar if you have used other coding agent CLIs. A limited-time free usage offer is available in both Grok Build and Cursor (x.ai/cli).
In Cursor, Grok 4.5 is available on all plans. It was trained alongside Cursor — the training used real coding data from the Cursor editor — which gives it a practical edge on the IDE-style workflows Cursor users already run (Cursor blog).
If you are comparing Grok 4.5 to other coding agents, the landscape looks like this: you get a dedicated coding environment with web research and code execution in OpenAI's Codex cloud sandbox, or you can run Codex locally at zero cost via the open-source CLI. Grok 4.5's angle is that it sits inside your existing editor and your existing Office apps — no new tool to adopt, just a model upgrade.
How do you call the Grok 4.5 API?
The API model name is grok-4.5. It supports both the Responses API and Chat Completions (OpenAI-compatible). Here is the minimal setup from the official docs (xAI docs):
curl https://api.x.ai/v1/responses \
-H "Authorization: Bearer $XAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "grok-4.5",
"input": "Find and fix the bug, then explain it: function median(a){a.sort();return a[a.length/2]}"
}'
Get an API key at console.x.ai. The model is also available through gateways like OpenRouter, Vercel, Cloudflare, Snowflake, and Databricks Mosaic if you prefer not to manage keys directly.
One critical implementation note from the docs: set a prompt_cache_key (on the Responses API) or x-grok-conv-id header (on Chat Completions) to route conversation requests to the same server. Without it, you often pay full input price on a cache-cold server because cache hits are unreliable across different servers.
What is the reasoning effort setting and when should you change it?
Grok 4.5 supports configurable reasoning effort: low, medium, or high (high is the default). This controls how many reasoning tokens the model spends thinking before responding.
- High (default): Best for complex coding, multi-step agentic tasks, and problems requiring deep analysis. Uses more output tokens but delivers higher accuracy.
- Medium: Good for everyday questions, drafting, and tasks where speed matters more than maximum precision.
- Low: Fastest and cheapest. Use for simple lookups, formatting, or when you are iterating quickly and want rapid feedback.
The reasoning effort setting is the single biggest lever on cost per task. If your workflow involves many quick exchanges (like iterative refinement), dropping to medium or low can cut your token bill significantly without meaningfully reducing quality on simpler requests.
What does Grok 4.5's "end-to-end task completion" actually look like?
SpaceXAI's positioning is that Grok 4.5 does not just answer questions — it takes a task from start to finish. The announcement includes a demo where the model generated a full Three.js solar system simulation from a single prompt, with no build step and no follow-up instructions (xAI announcement).
On the office side, "end-to-end" looks like this: you hand Grok a research paper in PowerPoint, and it produces a nine-slide deck with native shapes. You ask it in Excel to build a financial model, and it creates multi-sheet formulas with live web research and explanatory sticky notes. You describe a landing page in Grok Build, and it returns a working single-file HTML page.
The long-context reasoning — 500K tokens — is what makes single-prompt generation viable. The model holds the entire codebase or document structure in context and keeps it consistent, rather than losing the thread halfway through a long file.
Is Grok 4.5 available in the EU?
Grok 4.5 launched outside the EU on July 8, 2026. EU availability arrived on July 17, 2026, when SpaceXAI opened the API console for EU users (xAI release notes). If you are in the EU and were blocked at launch, access is now open through the standard channels.
What this means for you
If you write code: Install Grok Build or open Cursor and start with the free trial. Set prompt_cache_key on API calls to lock in the cached-input discount. Drop reasoning effort to medium for iterative prompt-and-refine loops; bump it back to high for complex debugging or architecture decisions.
If you work in documents, spreadsheets, or decks: Install the Grok add-ins from the Microsoft AppSource marketplace (search "Grok" in Insert > Add-ins inside any Office app). You need a SuperGrok subscription ($30/month) or a Grok Business plan. The real-time X data feed is genuinely useful for research-heavy decks and competitive analysis — that is the ceiling of what a Copilot cannot do.
If you build AI applications: Grok 4.5 is the cheapest frontier API by token price, but model the full cost — tokens plus server-side tool calls (web search at $5/1k, code execution at $5/1k, file search at $10/1k). For agentic workflows that make many tool calls, the tool budget can exceed the token budget. Track cost_in_usd_ticks in the API response for ground-truth cost per request. If you want to keep your total AI spend low across multiple models, the delegate-to-free strategy of routing heavy lifting to free worker models applies here — use Grok 4.5 where its price-to-performance wins, and cheaper models where they do not.
If you want to learn prompting that works everywhere: The same model running across all surfaces means the prompting patterns you practice in Cursor or Grok Build transfer directly to Word, Excel, and X. Start with the surface closest to your daily work.
FAQ
Q: Is Grok 4.5 free to use? A: Grok 4.5 has a free tier inside X and grok.com (limited usage), a limited-time free trial in Grok Build and Cursor, and a pay-per-token API at $2/$6 per million tokens. Office add-ins require a paid SuperGrok ($30/month), Heavy ($300/month), Business ($30/user/month), or Enterprise plan.
Q: How does Grok 4.5 compare to ChatGPT and Claude? A: On published benchmarks, Grok 4.5 is competitive but trails Claude Opus 4.8 and GPT-5.5 on overall intelligence indices. Its edge is price (roughly 60-76% cheaper on API token costs) and token efficiency (4.2x fewer output tokens on SWE Bench Pro). Grok is the only model with native real-time X data access.
Q: Can I use Grok 4.5 without an X account? A: Yes. You can access Grok 4.5 through grok.com, the API console at console.x.ai, Cursor, Grok Build, and the Office add-ins — none of which require an X account. An X Premium+ subscription ($8/month) is just one of several access paths.
Q: What is the context window for Grok 4.5? A: 500,000 tokens. Requests exceeding 200K tokens in context are billed at a higher rate ($4 in / $12 out per million) per the xAI pricing page.
Q: Does Grok 4.5 work offline or locally? A: No. Grok 4.5 is a cloud-hosted model served from SpaceXAI's data centers. There is no local or self-hosted version. If you need a local frontier model, you can run a 744B parameter LLM locally with Colibri as an alternative.
Q: What subscription do I need for the Microsoft Office add-ins? A: The Office add-ins (Word, Excel, PowerPoint, Outlook) require a SuperGrok, SuperGrok Heavy, Grok Business, or Grok Enterprise plan. Usage limits apply. The add-ins are free to install from the Microsoft marketplace but will not function without a qualifying paid subscription.
Discussion
0 comments