Google’s latest AI release is a sprint in two directions: faster, cheaper Flash models for developers, and no sign of the bigger Gemini upgrade the company had promised.
The tension began at Google I/O in May, when the company said Gemini 3.5 Pro would arrive in June. June passed, and the flagship model remains absent; one account argues that customers invested in Google’s tools are being left with incremental Flash improvements instead.1
Three weeks after releasing Gemini 3.6 Flash, Google introduced Gemini 3.7 Flash as its replacement. The company says the new model is better at following instructions, planning through multistep tasks and using tools—upgrades aimed squarely at coding and knowledge work.2 It is rolling into the Gemini API, AI Studio and Enterprise, while Gemini Spark gives AI Pro and Ultra subscribers a more capable agent for complex Google Workspace workflows.3
Google’s pitch is velocity as much as capability. CEO Sundar Pichai called Flash models “workhorses that offer performance at a great price,” saying the company is shipping updates quickly to put them in developers’ hands.
4 The introductory rate—$0.75 per million input tokens and $3.75 per million output tokens—is half the initial price of 3.6 Flash.2
There is outside validation for the smaller model’s role. Perplexity CEO Aravind Srinivas said Flash models are “great for fast, cost-efficient subagents” and that his company uses them in its Computer harness.
5 Google-linked benchmark claims also point to gains: a repost highlighted a 65.5% DeepSWE result, up 18.8% from 3.6 Flash while costing less than half as much per task.6
But the release also sharpens the uncomfortable question. Critics see a rapid-fire Flash cadence as a way to sustain momentum while Google avoids putting a delayed 3.5 Pro beside the newest offerings from OpenAI and Anthropic. Google has not explained the model’s fate, even as it says work toward Gemini 4 is showing promising early results.2