A Speed-Tier Model That Now Plays in Frontier AI Territory
Gemini 3.5 Flash marks an inflection point in Google’s AI strategy. Long framed as the lightweight counterpart to Pro-tier systems, the new Flash model lands squarely in frontier AI performance territory. On the independent Artificial Analysis Intelligence Index, it posts a composite score of 55, coming within two points of Anthropic’s Claude Opus 4.7 and five points of GPT-5.5. That puts a speed‑optimized model in striking distance of the very systems that have dominated benchmark leaderboards in recent months. Crucially, this performance is coupled with aggressive pricing. Gemini 3.5 Flash is listed at USD 1.50 (approx. RM6.90) per million input tokens and USD 9.00 (approx. RM41.40) per million output tokens, while GPT-5.5 launched at USD 5.00 (approx. RM23.00) input and USD 30.00 (approx. RM138.00) output. Combined with its strong benchmark showing, Flash forces a re-think of how much “frontier-level” intelligence should cost.
Beating Google’s Own Pro Model on Coding and Agentic Tasks
Perhaps the most telling comparison is not with rivals, but with Google’s own lineup. Gemini 3.5 Flash, officially a speed-tier model, outperforms the flagship Gemini 3.1 Pro across several coding benchmark results and agentic tests. On Terminal-Bench 2.1, a coding benchmark, Flash scores 76.2% versus 3.1 Pro’s 70.3%. For real-world agentic AI capabilities measured by GDPval-AA Elo, Flash reaches 1656, far ahead of Pro’s 1314. It also leads in scaled tool use, scoring 83.6% on MCP Atlas compared with 78.2% for 3.1 Pro. These numbers undercut the traditional assumption that cheaper, latency-optimized models must lag substantially in capability. Google describes 3.5 Flash as delivering intelligence that rivals large flagship models on multiple dimensions, including multimodal understanding, where it records 84.2% on the CharXiv Reasoning benchmark. In practical terms, developers can now reach higher levels of coding automation and tool orchestration without moving up to a Pro-tier system.
Fourfold Speed Gains and a New Cost Curve for Frontier AI
Gemini 3.5 Flash does not just aim to be smarter; it is engineered to be dramatically faster. Google CEO Sundar Pichai said the model delivers around 289 tokens per second, with Google and partner reports characterizing it as four times faster than comparable frontier AI models. From an infrastructure perspective, that kind of throughput can reshape the economics of running large fleets of agents, enabling more experimentation and sustained workflows at lower latency. Pricing magnifies this effect. Flash’s listed rates of USD 1.50 (approx. RM6.90) per million input tokens and USD 9.00 (approx. RM41.40) per million output tokens compare against GPT-5.5’s USD 5.00 (approx. RM23.00) input and USD 30.00 (approx. RM138.00) output. That places Flash at roughly one-third the cost of some competing frontier offerings, while staying within a few points of their benchmark scores. For enterprises weighing large-scale deployments, the AI model pricing comparison is no longer a minor detail; it becomes a core strategic lever.
Built to Act: Long-Horizon Agents, Not Just Chatbots
Where earlier Gemini releases leaned heavily on conversational prowess, 3.5 Flash is explicitly designed for action-oriented, long-horizon tasks. Google positions it as its strongest agentic and coding model yet, optimized for workflows in which AI must plan, build, and iterate across many steps. With support for deploying multiple subagents in parallel through platforms such as Google’s Antigravity, the model can coordinate complex pipelines, from navigating large codebases to orchestrating multi-step business processes. The elevated GDPval-AA score indicates a significant leap in real-world agent performance, and Google notes that partners in sectors like banking and fintech have used 3.5 Flash to compress multi-week workflows into far shorter cycles. Under supervision, it can reliably execute multi-step coding tasks and tool-based operations while sustaining frontier AI performance. This shift from answering to acting signals where Google sees the next wave of value: AI systems that operate as persistent digital workers rather than passive assistants.
Google Makes Gemini 3.5 Flash Its New Default AI Model
Google is backing its confidence in Gemini 3.5 Flash by making it the new default across core products. The model now powers the Gemini app and AI Mode in Google Search, effectively placing a frontier-adjacent system in front of mainstream users by default. It also serves as the engine behind Gemini Spark, a personal AI agent that runs continuously to take actions on a user’s behalf, currently rolling out to testers. A more capable 3.5 Pro model is promised next month, targeting deeper reasoning and high-context tasks. Yet Flash’s blend of frontier AI performance, cost efficiency, and speed already complicates the competitive landscape for Anthropic and OpenAI. If Google can maintain this trajectory—tightening the gap in intelligence while keeping prices lower—the economics of frontier-scale AI deployment could tilt in favor of fast, agent-first models like Gemini 3.5 Flash, rather than the most expensive flagships.