Gemini 3.5 Flash Lands - Google's Cheaper Tier Outperforms Its Own 3.1 Pro Flagship

Google DeepMind has released Gemini 3.5 Flash, the first model in its new 3.5 family that the company positions as combining frontier intelligence with practical applications. The release creates an unusual situation where Google's cheaper Flash tier now outperforms its own flagship 3.1 Pro model, which launched just months earlier in February. On Terminal-Bench 2.1, Gemini 3.5 Flash achieves a score of 76.2%, compared to 3.1 Pro's 70.3%. The performance gains extend beyond accuracy, with Flash delivering 289 tokens per second—roughly four times faster than competing frontier models—whilst the API costs approximately 40% less than the model it has just surpassed. Google's strategic focus has shifted markedly towards agentic capabilities rather than traditional chatbot interactions. The company emphasises Flash's ability to plan across large codebases, spawn subagents for parallel task execution, and maintain context over extended periods. This positioning suggests Google views the model as an autonomous coding teammate rather than a conversational assistant. The timing raises questions about Google's model hierarchy, particularly with Gemini 3.5 Pro delayed until June. If the Flash variant can already exceed the previous Pro model's performance at a fraction of the cost, industry observers will be watching closely to see what capabilities the Pro version delivers when it arrives next month.