What Is Grok 4.5?

xAI, now operating under the SpaceXAI brand, released Grok 4.5 on July 8, 2026. The company describes it as “SpaceXAI’s smartest model built for coding, agentic tasks, and knowledge work,” while Elon Musk positioned it as “an Opus-class model, but faster, more token-efficient and lower cost.”

Try Apidog today

The core specs are straightforward: $2 per million input tokens, $6 per million output tokens, a 500k context window, around 80 tokens per second, and training that included real developer sessions from Cursor, the AI code editor SpaceX agreed to acquire in June. Grok 4.5 is available in Grok Build, in Cursor on all plans, and through the xAI API.

This guide focuses on what developers need to know: where Grok 4.5 fits, what the published benchmarks actually show, when to use it, and how to call it from code. If you want to test the endpoint before adding it to your app, Apidog lets you send requests to grok-4.5, inspect streaming responses, and mock the API for your team without writing client code first.

Grok 4.5 at a glance

Spec	Value
Model ID	grok-4.5
Release date	July 8, 2026
Context window	500,000 tokens
Input price	$2.00 per 1M tokens
Output price	$6.00 per 1M tokens
Speed	~80 tokens/second
Availability	Grok Build, Cursor, xAI API
EU availability	Not yet; expected mid-July 2026
Endpoint	https://api.x.ai/v1/responses

Two implementation details matter most.

First, the context window is 500k tokens. That is large, but the older and cheaper grok-4.3 offers 1M tokens. If your use case is whole-repository analysis, long-document RAG, or large transcript processing, do not assume the newer model is automatically the better fit. Test both against your actual prompt sizes.

Second, EU availability is not immediate. xAI says Grok 4.5 “is not yet available in the EU in any SpaceXAI products or the API console,” with availability expected in mid-July 2026. If your application serves European users, plan for fallback behavior or region-based routing.

What changed in Grok 4.5

1. It was trained with Cursor usage data

The defining release detail is that Grok 4.5 “was trained alongside Cursor.” According to Cursor’s announcement, the editor contributed trillions of tokens from “user interactions with codebases and software tools.”

For developers, that means Grok 4.5 was not trained only on static source code. It also learned from workflow-like data such as:

• debugging traces
• multi-file edits
• tool interactions
• agent attempts and corrections
• developer feedback during real coding sessions

That should make the model especially relevant for agentic coding tasks, refactors, and iterative debugging workflows.

The approach also raises privacy and consent questions. For more context, see what Cursor session training means for developers.

2. Reinforcement learning targets multi-step engineering

xAI says Grok 4.5 was trained across tens of thousands of NVIDIA GB300 GPUs, with reinforcement learning applied to “hundreds of thousands of tasks, centered on multi-step software engineering.”

The stated goal is “per-token intelligence”: solving tasks with fewer, more useful tokens instead of long reasoning chains. That matters for API users because output tokens affect both latency and cost.

3. It targets office workflows too

Grok 4.5 is the default model in Grok Build, and xAI’s announcement also emphasizes productivity tasks outside code. The model is described as supporting:

• multi-sheet Excel models with web research
• PowerPoint diagrams built from native shapes
• Word document drafting through xAI Office plugins

Cursor frames Grok 4.5 as “the first we’ve built for more than software engineering.”

Read the benchmarks carefully

xAI published four coding benchmarks. Here is the full comparison shown in the release materials, including models that outperform Grok 4.5 on xAI’s own charts.

Benchmark	Grok 4.5	Opus 4.8 (max)	GPT 5.5 (xhigh)	Fable (max)
DeepSWE 1.0 (pass@1)	62.0%	55.75%	64.31%	66.1%
DeepSWE 1.1	53%	59%	67%	70%
Terminal Bench 2.1	83.3%	78.9%	83.4%	84.3%
SWE Bench Pro (resolve)	64.7%	69.2%	58.6%	80.4%

The practical takeaway: Grok 4.5 is competitive, but not the top model on every benchmark.

Against Claude Opus 4.8, the comparison is mixed. Grok 4.5 wins DeepSWE 1.0 and Terminal Bench 2.1, but loses DeepSWE 1.1 and SWE Bench Pro. “Opus-class” is reasonable. “Strictly better than Opus” is not supported by these numbers. For a deeper comparison, see Grok 4.5 vs Claude Opus 4.8.

Claude Fable 5 tops all four listed benchmarks on xAI’s own chart. That makes Grok 4.5 less of a peak-capability claim and more of a cost/speed/capability tradeoff.

Also note the provenance. xAI says competitor results come from published system cards or benchmark leaderboards, and that DeepSWE evaluations were created by Datacurve with runs handled by outside parties. That is more transparent than pure self-reporting, but it is not the same as a fully independent third-party evaluation of Grok 4.5. The benchmark details are tracked in this Grok 4.5 benchmarks deep dive.

Token efficiency is the strongest reason to test it

The most interesting number is not a benchmark score. It is output token volume.

On SWE Bench Pro, Grok 4.5 resolves tasks using an average of 15,954 output tokens. Opus 4.8 uses 67,020 output tokens on the same benchmark, roughly 4.2x more.

That affects two things immediately:

1. Cost: output tokens are billed at $6 per million.
2. Latency: output tokens are generated at around 80 tokens per second.

For agentic workflows, a single task can include many prompts, tool calls, intermediate observations, and retries. If a model reaches a solution with fewer tokens, the real-world cost difference can be larger than the list price suggests.

xAI claims Grok 4.5 has “roughly 2x the token efficiency of comparable leading models” overall and solves tasks “in under half the number of steps.” You should validate that against your own workloads. See Grok 4.5 pricing explained for a cost breakdown.

How to start using Grok 4.5

You have four practical entry points.

1. Use Grok Build

Grok 4.5 is the default model in Grok Build. xAI says it is “offering free Grok 4.5 usage for a limited time.”

Use this path if you want to test prompts manually before building an integration.

2. Use Cursor

Grok 4.5 is available in Cursor on all plans across desktop, web, iOS, CLI, and SDK. Cursor also announced doubled usage for the first week.

If your use case is coding assistance, refactoring, or repo-level agent workflows, Cursor is the fastest way to evaluate the model in a real development environment. Setup steps are covered in the Cursor guide.

3. Use Grok CLI

Install the CLI with:

curl -fsSL https://x.ai/cli/install.sh | bash

Use this option for quick terminal-based experiments, shell workflows, and local automation.

4. Call the xAI API

For application integration, use the Responses endpoint:

curl -s https://api.x.ai/v1/responses \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-4.5",
    "input": "Summarize the breaking changes in this OpenAPI diff."
  }'

The model ID is important: use grok-4.5 with a dot. grok-4-5 is not the same identifier.

A full walkthrough with streaming and Apidog-based testing is available in how to use the Grok 4.5 API. If you want to avoid spending API credits while experimenting, see how to use Grok 4.5 for free.

Test the API workflow with Apidog

Once you can call the endpoint, use Apidog to make the integration repeatable.

A practical workflow:

1. Create a new request for https://api.x.ai/v1/responses.
2. Store your API key as an environment variable.
3. Add the Authorization header:

Authorization: Bearer {{XAI_API_KEY}}

1. Add a JSON body:

{
  "model": "grok-4.5",
  "input": "Summarize the breaking changes in this OpenAPI diff."
}

1. Save test cases for your common prompts.
2. Compare grok-4.5 against grok-4.3 using the same inputs.
3. Mock the expected response shape so frontend work can continue while prompts are still being tuned.

You can download Apidog free to follow along.

Where Grok 4.5 fits in the 2026 model landscape

Grok 4.5 sits in a crowded mid-frontier tier. It is priced below Claude Opus 4.8, listed at $5 input and $25 output per million tokens, and far below Claude Fable 5. It also scores within a few points of Opus on several coding benchmarks.

Against open-weight models such as GLM 5.2, Grok 4.5 wins the published coding evaluations but costs more per token.

The practical pitch is not “best model overall.” It is:

• lower cost than some frontier competitors
• fast generation
• strong coding benchmark performance
• better token efficiency for agentic tasks

If your bottleneck is agent throughput or API spend, Grok 4.5 is worth benchmarking. If you need the highest score on the hardest reasoning tasks, xAI’s own charts point to other models.

For more background, see what is Claude Opus 4.8 and the earlier coverage of Grok 4.1.

FAQ

Is Grok 4.5 free to use?

For a limited time, yes, in Grok Build and Cursor. API usage is paid at $2 per million input tokens and $6 per million output tokens. Current free options are listed in the free-access guide.

What is the Grok 4.5 model ID?

The model ID is:

grok-4.5

Use the dot. grok-4-5 will return a model-not-found error.

What is the Grok 4.5 context window?

Grok 4.5 supports a 500,000-token context window. The older grok-4.3 supports 1M tokens, so choose based on workload rather than release date.

Is Grok 4.5 better than Claude Opus 4.8?

Not across the board. It wins two of the four benchmarks xAI published and loses the other two. It is faster and cheaper, but the published data does not show it as stronger on every task.

Can I use Grok 4.5 in the EU?

Not yet. xAI says EU availability across products and the API console is expected in mid-July 2026.