跪拜 Guibai
← All articles
Cursor · AI Coding · Grok

Grok 4.5 Lands in Cursor, Undercutting Claude on Price for Coding

By 卡卡罗特AI ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

A top-tier coding model at 1/17th the price of the nearest competitor changes the unit economics of AI-assisted development, especially for teams that don't need absolute peak reasoning. Cursor's integration also means Grok 4.5 slots into an existing editor workflow rather than requiring a separate tool.

Summary

Grok 4.5 is the first model SpaceXAI has positioned explicitly for coding and agent workflows, and it was trained jointly with Cursor. Third-party benchmarks place it fourth in overall AI capability and second in programming, ahead of Claude Opus 4.8. Musk himself acknowledged the model trails Fable5 in raw performance but argued most real-world tasks don't need that ceiling.

Pricing is the headline: a community comparison chart shows Grok 4.5 scoring close to Fable5 while costing about 1/17th as much. Access comes through two channels: a $30/month CLI agent called Grok Build, or a Cursor Pro subscription starting at $20/month, which also bundles GPT-5.5 and multiple Claude models under a shared usage quota.

Early demos include a rocket-tracking app with a 3D Earth view, an interactive solar-system simulation from xAI, and a physics-based 3D flag animation with realistic cloth simulation. The model is available now inside Cursor with Alipay payment support.

Takeaways
Grok 4.5 is the first model SpaceXAI has marketed as purpose-built for programming and agent tasks.
Third-party rankings put it fourth overall and second in coding, ahead of Claude Opus 4.8.
Pricing lands at roughly 1/17th of Claude Fable5 for comparable benchmark scores.
Cursor co-trained the model and offers it inside a $20/month Pro plan alongside GPT-5.5 and multiple Claude versions.
A standalone CLI agent, Grok Build, costs $30/month with no visual interface.
Cursor subscriptions accept Alipay, lowering the payment barrier for Chinese developers.
Early demos show real-time 3D simulations and physics-based animations generated directly from prompts.
Conclusions

SpaceXAI is explicitly conceding the absolute quality crown to Anthropic and competing on cost instead, a strategy that mirrors the cloud-infrastructure playbook.

Co-training with Cursor suggests the model's programming strength may be tuned specifically for the editor's agent loop rather than general-purpose code generation.

Bundling Grok 4.5 inside Cursor alongside competing models turns the editor into a model marketplace where users arbitrage capability against quota consumption.

Concepts & terms
Grok Build
A CLI-based terminal agent tool from SpaceXAI, analogous to Claude Code, that provides direct access to Grok models without a graphical interface.
Cursor Pro membership
A paid subscription tier for the Cursor editor that unlocks access to multiple frontier models including GPT-5.5, Claude variants, and Grok 4.5, subject to usage quotas.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗