Version 3

Google launched Gemini 3.5 Flash at Google I/O 2026 on May 19, priced at $1.50/million input tokens and $9/million output tokens — 3x the cost of Gemini 3 Flash Preview and 6x Gemini 3.1 Flash-Lite [10][11]. The model replaced Gemini 3 as the default on Google's web interface and now powers Google Search AI Mode [7], while Google's official launch blog positioned it under the tagline 'frontier intelligence with action' [1]. Community and developer reaction to the pricing has been sharply negative, with one observer characterizing it as the worst negative feedback they had ever seen for a software release [20], while early benchmark data rates it a clear capability leader on the intelligence-vs-speed Pareto frontier [14].

Gemini 3.5 Flash breaks the developer expectation that 'Flash' means affordable — its API price nearly matches Gemini 3.1 Pro while being deployed for free in consumer products, creating a structurally confusing cost signal. The breadth and speed of the pricing backlash suggests this repricing is not being absorbed quietly, and continued migration-focused coverage [13] indicates developers are actively weighing alternatives rather than simply accepting the new tier.

Google announced Gemini 3.5 Flash at Google I/O 2026 on May 19, marketing it under the tagline 'frontier intelligence with action' [1][2]. The model was spotted in the Google Cloud Console hours before the official announcement [3][4][5], alongside a companion model called Gemini Omni Flash [6]. Upon launch, Google confirmed that Gemini 3.5 Flash had replaced Gemini 3 as the default model on its web interface, that it was powering AI Mode in Google Search [7], and that it featured in updated Google AI subscription offerings announced at I/O [8]. The Gemini API changelog [9] and Google's official model blog [1] formalized its release documentation.

The pricing landed at $1.50 per million input tokens and $9 per million output tokens — 3x the cost of Gemini 3 Flash Preview and 6x Gemini 3.1 Flash-Lite [10][11]. One analysis noted the output token price is nearly 30x more expensive than the previous cheapest Flash-tier option [12]. An independent benchmark run illustrated real-world cost impact: the Artificial Analysis benchmark cost $1,551.60 on Gemini 3.5 Flash (high), compared to $892.28 for Gemini 3.1 Pro Preview [10]. A dedicated pricing migration comparison [13] confirmed developers are actively researching the cost jump and their options. These figures nearly erase the price gap between Google's mid-tier and pro offerings, fundamentally undermining the traditional cost/capability segmentation that the 'Flash' label had previously signaled to developers.

On the capability side, early signals are more favorable. Artificial Analysis rated Gemini 3.5 Flash as the 'clear leader on the Intelligence vs Speed Pareto frontier' with large gains on its benchmarks [14]. An early-access user described first impressions as feeling 'on par with Gemini 3.1 Pro Preview' [15], and a French-language commentator noted it was 'more performant and less expensive than the current Pro version' [16]. These assessments suggest the model does represent a genuine capability step up from prior Flash offerings — but the question of whether that justifies prices matching the Pro tier remains contested.

The developer community's pricing reaction was swift and unusually intense. Across social media and technical forums, posts described the pricing as 'painful' [17], announced 'the good times are over' [18], and declared outright that 'Gemini 3.5 Flash is EXPENSIVE' [19]. One observer stated they had 'never seen such negative feedback in a software release before' when characterizing the response [20]. Analyst Simon Willison contextualized this as part of a broader industry repricing trend, noting that Google, OpenAI, and Anthropic all appear to be simultaneously testing API customer price tolerance — suggesting the aggressive discounting of early LLM competition may be giving way to margin recapture across the industry [10]. On the tooling side, the llm CLI plugin for Gemini added gemini-3.5-flash support in version 0.32 [21], and a concurrent alpha (0.32a0) introduced streaming reasoning token support [22].