<p>Three frontier AI models dropped within nine days of each other this month, which is either the healthiest competition tech has seen in years or a sign everyone's compute bill is due at the same time. Google's entry showed up fashionably late — six weeks behind schedule — but it brought a genuinely absurd party trick.</p>

<h2>Two Million Tokens, One Very Expensive Catch</h2>
<p>Google DeepMind released Gemini 3.5 Pro on July 17, and the headline spec is a 2-million-token context window — double anything else currently shipping from a major lab. That's large enough to swallow an entire codebase, a stack of legal contracts, or your group chat's entire multi-year history in one prompt, no retrieval tricks required.</p>
<p>The catch: Deep Think, the model's extended-reasoning mode built for genuinely hard problems, is locked behind the $250-a-month Ultra subscription. The model itself lands a week after GPT-5.6 and nine days after Grok 4.5, making July the most crowded month of frontier-model releases so far this year.</p>

<h2>The Context Window Arms Race Gets Real</h2>
<p>A 2-million-token window isn't just a bigger number on a spec sheet — it changes what you can actually hand the model without chunking, summarizing, or praying your vector database indexed the right paragraph. For developers and researchers drowning in RAG pipelines, that's the difference between wrangling infrastructure and just... pasting the whole thing in.</p>
<p>But gating the best reasoning behind a premium tier is the tell everyone should be watching. Frontier labs are quietly admitting that their most powerful thinking modes cost real money to run, and the free-and-cheap tiers are increasingly a taste, not the whole meal.</p>

<p>Bigger context, smaller free lunch — that's the actual headline here, even if it doesn't fit as neatly on a slide.</p>

<p><em>Source: <a href="https://www.techtimes.com/articles/317919/20260606/google-gemini-35-pro… Times</a></em></p>