Full story
Gemini 3.8 Flash Arrives
Google added Gemini 3.8 Flash to AI Mode’s model menu, with Robby Stein, VP of Product for Google Search, announcing the rollout on X that 3.8 Flash is “available today in AI Mode for Google AI Pro & Ultra subs around the world.”
“3.8 Flash is “available today in AI Mode for Google AI Pro & Ultra subs around the world.””
The Search Engine Journal said the model you can select in AI Mode changes on Google’s model schedule, and that 3.8 Flash appeared in the “Gemini 3 models” section under “Ask anything” between “Auto” and “Pro.”
The same reporting described a rapid cadence, noting that Gemini 3.7 Flash joined the menu on August 14 and that 3.8 Flash is Google’s third Flash release in six weeks.
In parallel, Google’s own announcement framed Gemini 3.8 Flash as “our best reasoning & coding model yet,” while keeping the same introductory price as 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens.
The rollout also came with a cybersecurity variant, as Google introduced Gemini 3.8 Flash Cyber alongside the main model on Wednesday, September 2, 2026 local time.
Benchmarks, Costs, and Agents
Google described Gemini 3.8 Flash as its “most powerful reasoning and coding model to date,” and said both models continuously evaluate and optimize task execution results through long-running agent loops.
The Asian-language report said independent benchmarks showed the 3.8 Flash intelligence score rising to 59, while high-reasoning tasks incur costs 40% higher than the previous generation.
That same report put the introductory pricing at $0.75 per million input tokens and $3.75 per million output tokens, valid through the end of the year, and said the promotional pricing matched 3.7 Flash.
TradingKey highlighted DeepSWE v1.1 results, saying Gemini 3.8 Flash scored 73.7% versus Gemini 3.7 Flash’s 65.3%, and it also cited a CyberGym score of 86.2% for Gemini 3.8 Flash Cyber.
The Decoder added that Google says the performance gains come partly because the model “works harder,” executing extra reasoning steps and calling tools iteratively, which it linked to higher token consumption even when per-token prices stay the same.
Fairwind Access and Next Moves
Google said Gemini 3.8 Flash Cyber is provided to trusted defenders through the Fairwind Program, and the Asian-language report described Fairwind as granting over 650 global partner institutions “priority access” to cybersecurity AI solutions.
“available to trusted defenders through our new Fairwind Program”
The blog.google post said Gemini 3.8 Flash Cyber is “available to trusted defenders through our new Fairwind Program,” and it described the model as powered by long-running agentic loops that recursively evaluate and refine underlying models.
The Search Engine Journal reported that free-tier AI Mode users don’t get a model choice and instead get the default, while its model menu support page still lists only “Fast” and “Pro,” unchanged since 3.7 Flash arrived on August 14.
In the same thread of access and scheduling, the Search Engine Journal said the next thing to watch is whether 3.8 Flash moves from an option to the default, because a default change would reach the free tier.
Quartz also tied the release to Google’s broader AI lineup, noting that Chief Executive Sundar Pichai had said in May that a Pro model would arrive “next month,” but none has materialized, while Gemini 4 has yet to finish the posttraining phase.