Google just dropped Gemini 3.6 Flash, and the headline number is deceptively simple: 49% on DeepSWE, up from 37% for its predecessor. But the real story is not just the score jump , it's that the model achieved it while spending dramatically fewer tokens to get there. That combination is rare, and it changes the math for anyone running coding agents at scale.
What just landed on the leaderboard
Google launched three new Gemini models aimed at making AI agents faster, cheaper, and more efficient. The lineup includes Gemini 3.6 Flash, Gemini 3.5 Flash Lite, and Gemini 3.5 Flash Cyber. Gemini 3.6 Flash is the flagship of the three, and it's the one that matters most for coding agent workloads.
The model scored 49% on DeepSWE, compared with 37% for Gemini 3.5 Flash, reflecting higher precision and fewer unnecessary code changes. That's a 12-point absolute improvement , significant on a benchmark designed specifically to be hard to game.
What DeepSWE actually measures
To understand why this matters, you need to know what DeepSWE is testing. Most public coding benchmarks are starting to saturate , top models cluster within a narrow score band where the differences are inside the margin of error. DeepSWE was built to fix that.
- Contamination-free: Tasks are written from scratch, not adapted from existing commits or PRs, so no model has seen the solution during pretraining.
- High diversity: 113 tasks across 91 repositories in 5 languages (TypeScript, Go, Python, JavaScript, Rust).
- Real-world complexity:
Don't miss what's next in AI
Join 300,000+ engineers and researchers who get the signal, not the noise.
- Full access to in-depth AI research breakdowns
- Be the first to know what's trending before it hits mainstream
- Daily curated papers, repos, and industry moves