Jul 09, 2026 11:15:00

'Grok 4.5' has been released, offering performance equivalent to Claude Opus 4.8 at a lower price.

SpaceXAI announced its AI model ' Grok 4.5 ' on July 8, 2026, local time. Grok 4.5 is a model specialized for coding and agent tasks, and while it is less expensive than the cutting-edge models from OpenAI and Anthropic, it achieves benchmark scores equivalent to Claude Opus 4.8.

Introducing Grok 4.5 | SpaceXAI

https://x.ai/news/grok-4-5

Grok 4.5 | SpaceXAI Docs
https://docs.x.ai/developers/models/grok-4.5

Announcing Grok 4.5, our first model trained specifically for coding and agents. It was trained with Cursor and offers frontier intelligence at leading speeds and cost efficiency. https://t.co/i8HpU7w64k pic.twitter.com/oBjGtTsoNc

— SpaceXAI (@SpaceXAI) July 8, 2026

Grok 4.5 is a model developed in collaboration with Cursor, a subsidiary of SpaceX. It supports text and image input and can output text. The context window has a limit of 500,000 tokens. SpaceX CEO Elon Musk described Grok 4.5 as 'a base model with 1.5 trillion parameters that has been further trained with Cursor data.'

The following are the benchmark results for DeepSWE 1.0 , which measures the performance of AI models as coding agents. Grok 4.5 outperforms Claude Opus 4.8.

On the other hand, with DeepSWE 1.1, Claude Opus 4.8 scored higher.

In intelligence performance tests conducted by a third-party organization,

Artificial Analysis, it was rated as having performance second only to Claude Opus 4.8 and GPT-5.5, and surpassed Claude Sonnet 5 and GLM-5.2.

Grok 4.5 allows you to develop highly functional applications with minimal instructions.

The Grok 4.5 announcement page includes an embedded 'working solar system simulation' created with the short instructions: 'Create a beautiful simulation of space and the solar system. Need a fast simulation with adjustable time, realistic movement, orbits, stars, etc. Use three.js. HUD adheres to stylish and modern design principles.'

It can perform not only coding but also administrative tasks such as creating slides with high precision.

It also boasts higher token efficiency compared to other manufacturers' models. When running tasks in the coding benchmark SWE Bench Pro, Grok 4.5 produced an average of 1/4.2 times the number of output tokens per task compared to Claude Opus 4.8.

The API fee for Grok 4.5 is $2 (approximately 325 yen) per million tokens for input and $6 (approximately 975 yen) for output. The following graph shows the cost efficiency of various AI models, with the horizontal axis representing the cost per task and the vertical axis representing the intelligence score. It can be seen that Grok 4.5 has a lower cost than AI models with equivalent scores and a higher score compared to models with equivalent costs.

Grok 4.5 is available with all Grok Build and Cursor coding agent plans. It can also be run on other systems using the API. However, the release in the EU will be delayed until mid-July 2026.

Grok 4.5 was developed using the NVIDIA GB300 NVL72, and NVIDIA posted a comment upon the announcement of Grok 4.5 stating that it was 'the result of world-class AI infrastructure meeting world-class AI research.'

Congrats to @SpaceXAI on Grok 4.5 — trained on NVIDIA GB300 NVL72 systems and purpose-built for coding, agentic tasks, and knowledge work. This is what happens when world-class AI infrastructure meets world-class AI research. https://t.co/ZoWqE2v2sX

— NVIDIA (@nvidia) July 8, 2026