Near-Frontier Performance at a Discount
Gemini 3.5 Flash is reshaping AI model economics by pushing near-frontier scores into a lower-priced tier. On the Artificial Analysis Intelligence Index, it posts a composite score of 55, placing it within two points of Anthropic’s Claude Opus 4.7 and five points of GPT-5.5. Historically, Google’s Flash line has been positioned as a lighter, cheaper counterpart to Pro models, but Gemini 3.5 Flash blurs that hierarchy by delivering benchmark results that previously belonged to flagship systems. The model’s introduction also signals Google’s intent to compete directly with Anthropic and OpenAI on value rather than just headline intelligence. By making Gemini 3.5 Flash available in AI Mode in Google Search, the company is betting that a strong balance of capability and cost will entice both developers and end users, especially those sensitive to the total price of large-scale deployments.
Cost Per Token Efficiency Reframes AI Budgets
The economics behind Gemini 3.5 Flash may be its most disruptive feature. Google prices the model at USD 1.50 (approx. RM6.90) per million input tokens and USD 9.00 (approx. RM41.40) per million output tokens, roughly one third of GPT-5.5’s USD 5.00 (approx. RM23.00) input and USD 30.00 (approx. RM138.00) output rates. That cost per token efficiency means enterprises can approach frontier-level benchmark scores without paying frontier premiums. On the Artificial Analysis Intelligence Index, Gemini 3.5 Flash delivers its 55-point score at significantly lower total token costs than many peers, shifting how teams might structure model mixes across workloads. For use cases where slightly lower raw intelligence is acceptable, the model’s pricing offers a compelling tradeoff. It also hints at a broader strategy: Google appears willing to squeeze margins to win volume, putting pressure on competitors who have recently raised prices for their flagship APIs.
Speed, Benchmarks, and the New Performance Frontier
In AI model performance comparisons, Gemini 3.5 Flash doesn’t just compete on intelligence scores; it also excels on speed. It ranks fifth on the Artificial Analysis Intelligence Index, behind GPT-5.5, Claude Opus 4.7, Gemini 3.1 Pro Preview, and GPT-5.4, yet delivers faster inference than many of these models. Artificial Analysis reports that Gemini 3.5 Flash generates more than 280 output tokens per second, roughly 70% faster than Gemini 3 Flash and outpacing GPT-5.4 mini on the speed-intelligence frontier. On scatter plots of intelligence versus throughput, it resides in the attractive upper-right quadrant, combining high capability with high speed. This positioning is crucial for latency-sensitive applications such as interactive agents, code assistants, and real-time multimodal systems. By offering near-frontier intelligence at higher throughput, Gemini 3.5 Flash gives developers a practical alternative to heavier, slower, and more expensive frontier AI models.
Agentic Gains and Google’s Value-First Strategy
Beyond raw benchmarks, Gemini 3.5 Flash reflects Google’s push to strengthen agentic behavior while still competing on value. On GDPval-AA, a benchmark for economically valuable real-world tasks, it achieves an Elo score of 1,656—well ahead of Gemini 3 Flash and Gemini 3.1 Pro, and just behind GPT-5.4 (xhigh). It also shows notable improvements in tool use and long-horizon planning, even if this comes with longer multi-turn workflows. At the same time, Google emphasizes multimodal strengths and reduced hallucination rates, positioning Flash as a workhorse for complex, agentic applications rather than a simple budget option. With Gemini 3.5 Pro scheduled to follow, Flash effectively sets the floor of Google’s ecosystem closer to prior-generation frontier AI models. The result is a portfolio where value and speed, not just peak benchmark scores, define competitive advantage—potentially reshaping how enterprises evaluate and deploy frontier AI.