Google’s Gemini 3.7 Flash hits 65% DeepSWE, promising faster, cheaper coding agents
Gemini 3.7 Flash debuted with a 65.3% DeepSWE v1.1 score, up from 49% on 3.6 Flash, and better handling of stalled code‑agent loops. It adds three “thinking_level” modes and launches at half the prior API price—though rates double in 2027—so teams can test higher‑quality agents without breaking the bank.