Gemini 3.5 Flash Disappoints on Coding; 3.5 Pro Delayed One Month
Gemini 3.5 Flash, released at Google I/O as Google's intended workhorse coding model, scored approximately 50% on Cursor Bench — 15 percentage points below frontier models while costing roughly four times more per task than Cursor Composer 2.5. Sundar Pichai, interviewed at Google I/O, framed the workhorse model focus as mission-aligned with the requirement to serve billions of users without losing money per query. Gemini 3.5 Pro has been delayed by an additional month from its previously expected release window.
Why It Matters
A 15-point benchmark gap against a $0.50/task competitor at 4x the cost is a significant miss for Google's flagship workhorse positioning. The Pro delay further closes the window for Google to regain coding model competitiveness before enterprise decisions solidify around Cursor Composer 2.5.