Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
- Source
- deepmind.google
- Date
Google released Gemini 3.8 Flash with claimed improvements in software engineering, agentic tasks and multi-step reasoning while keeping the introductory token rates of Gemini 3.7 Flash. It also introduced Gemini 3.8 Flash Cyber for vulnerability discovery and automated patching through a restricted partner program. Artificial Analysis measured a three-point gain on its Intelligence Index, but also found 30 percent more output tokens and roughly 40 percent higher cost per task. That distinction matters for agent builders because a stable token rate can hide a larger total bill when the model reasons or acts for longer. Evaluate the model on completed workflows and latency, not the rate card alone.
Context
Google DeepMind attributes part of Gemini 3.8 Flash's stronger reasoning and coding performance to spending more effort on difficult tasks: taking additional reasoning steps and repeatedly calling tools. That means an unchanged price per does not necessarily produce an unchanged cost per completed task. The September 2 announcement says higher effort can consume more tokens and offers lower effort settings for applications constrained by compute. The useful comparison is therefore task success together with the total work needed to obtain it. Model announcement.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
Checking sign-in…
Loading comments…