Google updates Gemini Flash with an emphasis on deeper reasoning

The new model is designed to handle complex tasks by executing more reasoning steps and leveraging iterative tool calling. According to Google, these architectural shifts yield significant gains in software engineering and autonomous agent workflows. In internal benchmarks, Gemini 3.8 Flash outperformed both its predecessor and rival models like Anthropic’s Fable 5 on the DeepSWE v1.1, Vals Finance Agent V2, and Harvey’s Legal Agent tests.

Early performance analysis suggests that users should expect higher consumption patterns. Artificial Analysis noted that while per-token pricing is unchanged, the model’s increased output and agentic behavior resulted in a roughly 40% rise in cost per task during initial testing. Developers aiming to keep expenses predictable retain the option to stick with Gemini 3.7 Flash. Industry observers, such as Aigora.ai CEO John Ennis, view the trade-off favorably, noting that the model provides high-tier coding quality at speeds that open new possibilities for automated video production and complex agent tasks.