Skip to content
English
FikirPilot content

Google says its new Gemini 3.8 Flash model ‘works harder’ but could cost more

Updated: 5 Eyl 2026 · 2 min read · 375 words

Published: · Story reached us: · Processing time: 71 h 8 min

Google says its new Gemini 3.8 Flash model ‘works harder’ but could cost more
A modern data center corridor

Google introduced Gemini 3.8 Flash a few weeks after the release of its previous model. The company says the model takes more reasoning steps on complex tasks and uses tools iteratively. Its launch pricing is the same as Gemini 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens. However, Google warns that more tokens may be used to improve performance, particularly at higher effort levels. Developers seeking to reduce token consumption can continue using Gemini 3.7 Flash.

Artificial Analysis reported that the model is the cheapest option they have measured at this intelligence level, with costs approximately 40% higher than Gemini 3.7 Flash due to a 30% increase in output tokens per task and more processing rounds. Google announced significant improvements in software engineering and autonomous AI agents, saying the model outperformed its competitors in the DeepSWE v1.1, Vals Finance Agent V2, and Harvey’s Legal Agent tests. The model comes with safeguards designed to prevent misuse in the CBRN and cyberattack domains.

Gemini 3.8 Flash Cyber was launched under the Fairwind Program, which has 650 members and is open to governments and trusted partners. The program also provides access to the CodeMender tool. Gemini 3.8 Flash is available to consumers, developers, and enterprise users.

Why it matters

For developers, the key change is that while the same input price is maintained, the processing intensity of complex tasks may increase the total cost. Model selection therefore depends not only on the per-token price but also on how many output tokens and processing rounds the task requires; the previous version remains an option for those seeking lower consumption. The results from software engineering and autonomous agent tests provide a basis for comparison for teams evaluating the model’s use in these areas; however, some of the performance data appears to be based on Google’s own statements. Meanwhile, limiting access to the cybersecurity version to a restricted program sets its availability apart from general use and leaves open the question of who will be able to use this special version and under what conditions.

Background

Google is no new name in the FikirPilot archive: over the past 90 days, we have published 24 news stories mentioning the name; the latest is dated September 5, 2026.

Source: The Verge · Compiled by editor Altay Tuna Emir.