Most people will test Gemini 3.5 Flash the wrong way. [C001] It is worth watching even if you mostly use chatbots. The cheap part is agent execution, not chat turns. [C002]
If you only know AI through chat, this matters to you. A model update is only worth your time if it changes your next move. Judge this one by chat vibes alone, and you may spend time, budget, and attention on the wrong test.
Google's Gemini docs show the workflow stack around it: structured output plus Google Search, URL Context, Code Execution, File Search, and Function Calling. In plain English, one model can look things up, inspect files, use tools, and finish more of the job in one flow.
Google's own summary page leans into tool-use benchmarks, not conversation polish: 83.6% on MCP Atlas and 76.2% on Terminal-bench, versus Gemini 3 Flash at 62.0% and 58.0%. That is a strong clue about what Google thinks this model is for.
The pricing points the same way. Gemini 3.5 Flash Preview is listed at $0.50 per 1M input tokens and $3.00 per 1M output tokens. Batch pricing drops that to $0.25 and $1.50. Cheap chat is nice. Cheap multi-step execution is the real lever.
An update is worth watching only if it changes your next move, not because it ships a longer feature list. Test Gemini 3.5 Flash on a real tool-heavy task, not a chat demo. Boundary: this is based on Google's published summary page, pricing page, and tool docs, not my own lab run. If someone is judging it by chat vibes alone, send them this.