Breaking

SpaceXAI Releases Grok 4.6 for Long-Running Coding and Knowledge-Work Agents

August 12, 2026 at 11:49 EDT

Foundation Models
AI Agents
Software Dev & Coding

SpaceXAI released Grok 4.6 on August 12, positioning its new frontier model around long-running agents, coding, knowledge work, and more ambitious interactive and visual projects. The model is available through the SpaceXAI API, Grok Build, Cursor, OpenRouter, Vercel, and Cloudflare.

August 12, 2026 · SpaceXAI

Grok 4.6 goes long

The new frontier model is built to stay on task across extended coding, research, and visual-project workflows, with a 500,000-token context window.

500K

token context window

$2 / $6

per million input / output tokens

AA Intelligence Index

Coding benchmark · DeepSWE v1.1

A sizable gain, with room above

54%

Grok 4.5

65.9%

Grok 4.6

73%

GPT-5.6 Sol Max

Grok 4.6 gains 11.9 percentage points over its predecessor but trails the top shown score.

Long-horizon work

From idea to verified artifact

Research

→

Build

→

Self-test

→

Refine

What improved

Sustained execution

The model is trained for multi-step coding, knowledge work, visual projects, and checking its own output.

What remains open

Independent proof

The launch scores are company-reported, image output is unsupported, and long prompts carry higher token rates.

Continue reading

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