Breaking

SpaceXAI Launches Grok 4.7 as AI Coding Agents Move Beyond Toy Demos

September 25, 2026 at 09:46 EDT

Software Dev & Coding
AI Agents
Foundation Models

SpaceXAI has turned the AI-coding conversation toward long-running agents with Grok 4.7, a new model it describes as its strongest system for coding and knowledge work. The release lands as developers contrast today's agent workflows with a 2025 viral hexagon simulation in which Grok 3 visibly failed a simple-looking physics coding prompt.

Sep 21, 2026 | SpaceXAI

Grok 4.7 shifts coding from demos to agents

The new model is positioned for long-running coding and knowledge-work tasks, with stronger self-checking, a 500k context window, and lower base output pricing than several frontier peers.

500k

API context window

$2 / $6

input and output per 1M tokens

71.0%

DeepSWE v1.1 score

3.3%

risky cyber prompt pass-through

CursorBench lift

Grok 4.7 rises over Grok 4.6 on the same coding benchmark.

40.4%

Grok 4.6

46.3%

Grok 4.7

The agent loop

Plan->Use tools->Edit code->Verify

The benchmark target is shifting from a single generated snippet to repeated tool use across real projects.

What improves

Faster iteration, smaller task loops, stronger tool use, and better long-session follow-through.

What still matters

Verification overhead, almost-right code, review burden, and security boundaries remain central constraints.

Continue reading

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