Cursor Just Got a $60B Price Tag — Here's What the AI Models Powering Code Assistants Actually Cost

SpaceX's acquisition of Cursor parent Anysphere for $60 billion is all over the news today (WSJ, TechCrunch, CNBC all confirmed it), and it's a good reminder that while the valuations sound astronomical, the underlying AI inference costs are surprisingly manageable.

If you're building a coding assistant yourself, here's the honest picture. On the pricier end, gpt-5.5 runs $5.00 in / $30.00 out per million tokens — that output cost adds up fast in long code completions. Claude Sonnet 4 is a friendlier middle ground at $3.00 in / $15.00 out.

Want to go lean? DeepSeek V4 Flash is $0.14 in / $0.28 out — basically rounding error territory. DeepSeek V4 Pro steps it up a bit at $0.43 in / $0.87 out, and Kimi K2.5 sits at $0.57 in / $2.41 out. These are real models doing real work at costs that won't keep you up at night.

The middle tier — Gemini 3.1 Pro at $2.00 in / $12.00 out, or GLM-5.1 at $1.40 in / $4.40 out — gives you room to breathe without going bare-bones.

Sixty billion dollars buys a lot of vibes. A few cents per thousand tokens buys actual output.