Google’s newest coding model, Gemini 3.8 Flash, is tuned for long jobs and comes with an incredible launch discount.
Google shipped three Flash releases in six weeks, and the newest one is built for exactly the kind of work Junie does all day: multi-step engineering tasks that take real exploration to get right. Gemini 3.8 Flash is now available in Junie, in both the IDE plugin and Junie CLI, with a limited-time introductory offer of 75% off.
Slower per step, faster to a working answer
Gemini 3.7 Flash was built for quick, reliable answers. 3.8 Flash takes a different approach. It spends more tokens per task, exploring directories, retrying code that doesn’t work, and running verification steps before it calls a job done. Instead of generating a solution in one shot, it works in a loop: It breaks a goal into smaller sub-tasks, writes scripts to reproduce any bugs, runs tests through tool calls, and checks its own changes before finishing.
That shift shows up directly in Google’s own numbers on DeepSWE v1.1*, the benchmark built to test autonomous agents on long-horizon, real-world development work without data contamination. In testing, 3.8 Flash outperforms most larger frontier models while costing significantly less. The extra tokens aren’t wasted. On complex, multi-step work, they’re what turns a plausible-looking answer into one that’s actually been checked.
If your task just involves implementing a minor fix or small refactoring and doesn’t require the extra diligence of 3.8, 3.7 Flash is still available with full support and remains the cheaper choice per token.
75% off to start
For a limited time only, we’re offering Gemini 3.8 Flash at 75% off. Select it from the model picker in the IDE plugin or Junie CLI, and it’s ready to use, with no setup or waitlist required.
