Breaking

xAI Ships Grok 4.5, a Coding-Focused Model Co-Trained With Cursor

July 9, 2026 at 06:45 EDT

Foundation Models
AI Agents
Software Dev & Coding

xAI (SpaceXAI) formally released Grok 4.5, a new model built specifically for coding and agentic work, on July 8, 2026. It is the first model co-trained with the AI-native IDE Cursor, and is immediately available in Grok Build, all Cursor plans, and the SpaceXAI console (EU availability expected mid-July).

July 8, 2026 · xAI

Grok 4.5: the first model trained specifically for coding & agents

Co-trained with the Cursor IDE and built on a 1.5-trillion-parameter foundation, xAI's new release chases intelligence per dollar—half the price and twice the token efficiency of rivals.

$2/$6

per M tokens
input / output

500k

token context
window

~80

tokens / second
generation speed

1.5T

parameter V9
foundation model

Snorkel real professional tasks — pass rate

Higher is better. Bars drawn to scale.

29%

Grok 4.5

22%

GPT-5.5

21%

Opus 4.8

Harvey Legal Agent Benchmark

83.3%

Terminal-Bench

78%

SWE-Bench Multilingual

4th · 54

Artificial Analysis Intelligence Index

Efficiency is the pitch, not the top score

xAI frames Grok 4.5 around cost per outcome — completing the same task in fewer steps at a fraction of rival prices.

½ – ¼

price of competitors

≈ 2×

token efficiency

< ½

steps per task

Developers praise

High tool-calling reliability in multi-step agent loops
Strong output quality across long-running & cross-repo tasks
Token efficiency in Cursor and the Grok Build CLI

The caveat

Claude still leads on design, UI and creative work
Users likely to pick tools by use case
EU availability only expected mid-July

Available on day one

Grok Build → All Cursor plans → SpaceXAI console

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.