Breaking

xAI Releases Grok 4.5 With a Focus on Coding

July 9, 2026 at 02:56 EDT

Foundation Models
Software Dev & Coding
AI Agents

On July 8, 2026, xAI made its flagship model "Grok 4.5" generally available, tuning it for agentic coding that involves autonomous tool use and pushing it into the upper tier of independent benchmarks. Elon Musk had disclosed a private beta at SpaceX and Tesla in late June, and the general rollout has since been reported, including by Axios.

July 8, 2026 · xAI

Grok 4.5 goes GA — an efficient challenger for agentic coding

xAI's flagship model reaches the upper tier of independent benchmarks and ships across Grok Build, the API, and Cursor — tuned for autonomous tool use and priced to undercut rivals on token efficiency.

500k

token context length

Intelligence Index — 4th overall

~76

Coding Agent Index — on par with GPT-5.5

98.6%

pass on 69-task agent bench (~$0.84)

Token pricing — per 1M tokens

Same unit ($). Column height is proportional to price.

$2.00

Input

$6.00

Output

Low per-task token consumption points to a cost advantage versus rivals on comparable coding work.

How it improves: a real-workflow RL loop

Built on the V9 foundation model and tuned inside live engineering environments.

V9 base

Foundation model

→

Real dev data

Cursor + SpaceX/Tesla workflows

→

RL tuning

Continual improvement

→

Agentic coding

Autonomous tool use

Strengths seen in use

Instant access via Cursor across all plans
Reliable on long-running, tool-using tasks
End-to-end legal-doc workflow via Box MCP
Low token cost per task

Open questions

Production stability beyond benchmarks unverified
Real effectiveness of hallucination suppression
Part of Cursor's codebase possibly in training data
No EU availability yet

Continue reading

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