Grok 4.5 Cuts AI Coding Costs in Half – Threat to Anthropic and OpenAI

Why Grok 4.5’s Pricing Disrupts the AI Coding Market

SpaceX’s newest AI model, Grok 4.5, hit the market on Wednesday with a price tag that is roughly half of what Anthropic’s Claude Opus or OpenAI’s flagship agents charge. The claim – $2 per million input tokens and $6 per million output tokens – translates to a ↑ 90% reduction in cost per completed task, according to independent benchmarks.

The economics behind the launch

Artificial Analysis placed Grok 4.5 fourth on its real‑world agentic index, noting that it uses ↓ 50% fewer tokens than comparable models while delivering higher throughput. For enterprises that run autonomous agents for hours at a time, the token savings could shift the ROI equation dramatically.

“Our internal assessment is that Grok 4.5 is roughly comparable to Opus 4.7, but much faster,” Elon Musk wrote on X.

The model’s training draws on data harvested from Cursor’s AI‑first code editor, a platform that captured millions of developer interactions during the pandemic surge in remote coding. SpaceX’s Colossus supercomputer – touted at 200,000 Nvidia GPUs with a roadmap toward one million – supplied the compute horsepower needed for the upgrade.

Analysts at Reuters note that the AI coding market has coalesced around a few large players, with Anthropic holding roughly 50% share as of May 2026. The entry of a cost‑efficient alternative could force a re‑pricing wave, especially for firms that prioritize budget over marginal performance gains.

From a venture perspective, the $60 billion acquisition of Cursor is now being measured against Grok 4.5’s early traction. If the model can sustain its token‑efficiency in production, the deal may be reframed as a strategic data‑engineering play rather than a pure product purchase.

Industry sentiment remains mixed. Some developers praise the model’s ability to spin up a full‑stack rocket‑tracking app in minutes, while others caution that “vibes” – the intangible feel of reliability – will determine long‑term adoption.

Both Bloomberg and independent labs plan to publish deeper performance reports in the coming weeks, which should clarify whether Grok 4.5 can sustain its cost advantage without sacrificing the bug‑fix accuracy that enterprise teams demand.

Correction: An earlier dispatch misstated the token pricing; the correct rates are $2 per million input tokens and $6 per million output tokens.

Intel provided by Julian Reed (Consumer Electronics Expert).