Google introduces Gemini 3.6 Flash along with a competitor to Mythos.

      Google’s response to a summer of being outpaced is not to create a larger model, but rather to offer more affordable options. On Tuesday, the company introduced three new Gemini models, all in its fast, low-cost “Flash” tier, just a day before Alphabet announces its earnings. The emphasis is on efficiency rather than sheer power. However, there are still some notable exclusions.

      Affordable, quick, and widely available, yet not at the top

      The primary model is Gemini 3.6 Flash. Google claims it outperforms its predecessor in coding and knowledge tasks while utilizing approximately 17% fewer output tokens and at a lower cost—now $7.50 per million output tokens, down from $9. Its knowledge base will extend to March 2026.

      Next to it is 3.5 Flash-Lite, the fastest in the lineup at 350 tokens per second, and even more affordable. Both models capitalize on a key idea: most AI tasks do not require cutting-edge capabilities, just speed and affordability.

      Sundar Pichai, the chief executive, noted earlier this year that “companies are already rapidly exhausting their annual token budgets, and it’s only May.” He mentioned to Business Insider that a combination of Flash models could help businesses save over $1 billion annually.

      A more economical option for Mythos

      The most notable release is Gemini 3.5 Flash Cyber, which is specifically designed to detect and fix software vulnerabilities within Google’s CodeMender agent. Google describes it as a “cost-efficient” alternative to larger security models.

      Its unnamed competitor is Anthropic’s Mythos, which charges $10 per million input tokens and $50 per million output tokens, as reported by The Verge. Google asserts that Flash Cyber performs comparably to top-tier models on a significant benchmark at a fraction of the cost, directly challenging Anthropic’s edge in AI-driven security.

      However, as a vulnerability detection tool could also aid malicious actors, Google is implementing restrictions. Flash Cyber will only be available to governments and trusted partners in a limited pilot program.

      The model that did not launch

      There’s also the notable absence of Gemini 3.5 Pro, the flagship model Google had promised for June, which remains in testing, reportedly delayed due to underperformance in coding tasks. Currently, Google lacks a model within the public top ten.

      This delay is particularly stinging. In about a week, xAI’s Grok 4.5, three versions of OpenAI’s GPT-5.6, and Moonshot’s Kimi K3 have all been released. Meanwhile, Anthropic’s Fable 5 continues to top the leaderboards, according to reports from Reuters.

      Gemini 4, in theory

      Google’s response is to look to the future. The company announced that it has initiated its “most ambitious pre-training run yet” for Gemini 4. This is a declaration of intent rather than a functional capability.

      The push for efficiency goes deeper than just software. Google is also developing a custom chip to deliver Gemini at a lower cost. However, a crucial point to note is that all benchmarks cited are based on Google’s own evaluations, and no independent verification has been conducted yet.

      The strategy

      The plan appears coherent. In a year where companies are monitoring their token usage, offering affordable and fast options could capture the middle market while the flagship models catch up. The success of this strategy depends on two factors: the eventual release of 3.5 Pro and whether Gemini 4 proves to be more than just a training exercise.