Models & Access
Google introduces Gemini 3.8 Flash and restricted cyber access
Google pairs a general agent model with a trusted-defender variant; higher reasoning effort can increase token use despite unchanged introductory token rates.
What happened
Google announced Gemini 3.8 Flash for general reasoning and agent workflows, and Flash Cyber through its Fairwind trusted-defender programme. Google reports coding, reasoning and security gains. It says complex tasks may use additional reasoning steps and tool calls; the advertised introductory token pricing expires on 31 December 2026.
Why it matters for AI
A model’s token price does not determine the total cost of a completed task. Access restrictions also affect which workflows can actually use a capability.
AIRunsThis View
Our interpretation: compare completed, reviewed tasks under the available access tier. Include total tokens, retries, latency and intervention rather than ranking models by headline benchmarks.
What this could mean for us
A candidate for future workload evaluation, without an immediate platform switch or any assumption that restricted cyber access is available.
Source and method
This item is based on a primary source. AIRunsThis separates the source’s factual announcement from its own interpretation. Vendor performance claims have not been independently reproduced unless stated.