Gemini 3.8 Flash ships weeks after 3.7

Google shipped Gemini 3.8 Flash on September 2, its third Flash release in six weeks and the new default recommendation for software engineering, autonomous agents, and multi-step reasoning. The model scored 71% on DeepSWE v1.1, up from 65.3% for 3.7 Flash, and hit 89.4% on Terminal-Bench 2.1, edging past Claude Opus 5’s 89.1% on the same benchmark. It reads text, image, audio, video, and PDF input with a 1M-token context window and a 64K output cap.

Pricing holds at the 3.7 introductory rate through December 31: 75 cents per million input tokens, $3.75 per million on the output side, then both figures double to $1.50 and $7.50 once January 1, 2027 arrives. For anyone already routing work to Flash, this is a same-price upgrade worth taking immediately, with the price floor rising in four months a planning detail worth flagging now rather than at rollover.

Alongside it, Google released Gemini 3.8 Flash Cyber, a variant tuned for vulnerability detection and patch generation, distributed only through Fairwind, Google’s limited-access program for governments and vetted defenders. Google says it produced 2.6 times more correct Chrome vulnerability patches than larger commercial models in internal testing. That gated distribution puts it in the same critical-capability category OpenAI just crossed with Astra’s cyber threshold, where a model’s offensive-security skill outran its vendor’s comfort with open access. For teams weighing local versus cloud inference tradeoffs on cost and privacy, this release lands the same week as Perplexity’s hybrid on-device Mac mode, a reminder that the routing decision now spans both which model and where it runs. Anyone comparing Flash generations against the July release should also see the original Gemini 3.7 Flash launch for the benchmark baseline this update improves on.