Updated

A matte black cube with one bevelled edge catching the light on a dark background, abstract hero artwork for Grok 4.5

xAI

Grok 4.5

xAI's Grok 4.5, trained alongside Cursor for coding and office work, served at 80 tokens a second.

Chat with Grok 4.5 in Super AgentView pricing

What is Grok 4.5?

Grok 4.5 was xAI's smartest model at its July 16, 2026 launch, built for coding, agent tasks and knowledge work and trained alongside Cursor.

xAI says it is served at about 80 tokens per second and is roughly twice as token-efficient as comparable models, and reaches results in about half as many steps as other frontier models. It also handles Excel, PowerPoint and Word work, from multi-sheet models with web research to native PowerPoint diagrams. It takes text and images with a 500K-token context window. Grok 4.6 and Grok 4.7 have followed it.

Provider
xAI
Context window
500K tokens
Input
Text, Image
Reasoning
Low, medium, high
Knowledge cutoff
January 2026
Released
Jul 16, 2026

How Grok 4.5 performs

Published by xAI

Results xAI reported for Grok 4.5. Each lab runs its own setup, so compare these with other models only inside the same table.

Benchmark results xAI published for Grok 4.5
BenchmarkScoreSetting
SWE-Bench Pro64.7%high effort
SWE-Bench Multilingual78%high effort
Terminal-Bench 2.183.3%high effort
DeepSWE v1.153%high effort
APEX-SWE51.2%high effort
SWE-Marathon29%high effort
ProgramBench57.2%high effort
SWE-Atlas-QnA84%high effort
CAD-Bench88.1%high effort
Hallucination rate (lower is better)0.98%high effort, single turn

Source: media.x.ai

What you can do with Grok 4.5

  • Trained alongside Cursor on multi-step software engineering

  • About 80 tokens per second, according to xAI

  • About half as many steps as other frontier models, according to xAI

  • Handles Excel, PowerPoint and Word tasks

How to prompt Grok 4.5

  • Stay on medium for most tasks and use high for hard problems.
  • Ask for a spreadsheet layout or slide outline directly; office work is one of its stated strengths.
  • Try Grok 4.7 for long tasks that need self-checking.

Where Grok 4.5 fits best

  • Spreadsheet and slide planning

  • Fast coding help

  • Comparing Grok generations

  • Medium-effort everyday tasks

How Grok 4.5 compares

Grok 4.5 compared with other chat models on Flashloop
ModelLMArena textLMArena codingContext windowReasoning
Grok 4.51466#521514#48500K tokensLow, medium, high
GPT 5.6 Terra1467#491518#421.05M tokensOff to max
Claude Sonnet 51462#581520#371M tokensOff to max
Kimi K2.61461#591516#44256K tokensOn or off
GPT-6 Sol1459#621519#381.05M tokensOff to max

Good to know before you start

  • Grok 4.6 and Grok 4.7 are newer
  • Reasoning tops out at high and can't be turned off; xAI treats extra high as high on this model
  • Returns text only, and Super Agent uses Flashloop's generation models for any images or video

Start chatting with Grok 4.5

Chat with Grok 4.5 in Super Agent

  1. Select Grok 4.5

    Open Super Agent on Flashloop and choose Grok 4.5 in the model picker.

  2. Enter your prompt

    Ask for what you need: an answer, a plan, or the images and videos you want the agent to make.

  3. Send and keep going

    Send the message. Grok 4.5 answers in Super Agent and runs any generations you asked for.

Questions about Grok 4.5

Is Grok 4.5 still xAI's flagship?

No. It was xAI's flagship at launch in July 2026. Grok 4.6 and Grok 4.7 came after it.

What reasoning levels does Grok 4.5 support?

Low, medium and high. Super Agent starts it at medium.

What is Grok 4.5's context window?

500K tokens, with text and image input.

Where is Grok 4.5 in the model picker?

Under All models.

More chat models on Flashloop

Browse all models