Google Launches Gemini 3.5 Pro: 2-Million Token Context Window and Deep Think Reasoning Mode
After a one-month delay and a full base-model rebuild, Google has finally put Gemini 3.5 Pro on the table.
Three key facts
Gemini 3.5 Pro went live on July 17, headlined by a 2-million token context window. That is the largest production-grade context length among current frontier models — users can feed in an entire book, a full codebase, or hundreds of pages of research in a single request. Google had originally targeted a June launch, but the engineering team discovered structural failures in recursive tool-calling and SVG generation during internal testing, scrapped the base model, retrained from scratch, and pushed the release to today.
Deep Think reasoning mode is the other headline feature. This is Google's take on "slow thinking" — the model burns more reasoning steps before answering, trading latency and token cost for higher accuracy on complex problems. The feature is currently gated behind the $250/month Ultra subscription tier. On the API side, enterprise preview reports peg pricing at roughly $12–15 per million input tokens and $36–45 per million output tokens, with surcharges kicking in above 200K tokens.
The launch lands in the middle of a fierce model arms race. GPT-5.6 Sol and Grok 4.5 both went public on July 9; Claude Opus 4 shipped recently as well. The delay had put Google on the back foot, but the combination of a 2M context window and Deep Think carves out a differentiated position in long-document processing and complex reasoning — two verticals where sheer context length translates directly into capability.
WangDou's Take
Google did something surprisingly honest: when the base model wasn't good enough, they scrapped it and started over, eating a one-month delay instead of shipping half-baked. In an industry where "release the MVP and iterate" has become standard operating procedure, that actually stands out. But the 2-million token number, impressive as it sounds, comes with fine print — long-context surcharges on the enterprise API mean that actually maxing out the window could cost more per call than paying an intern to research for a day. Deep Think follows the same logic: more accurate but slower and pricier, essentially putting a price tag on the "compute for intelligence" trade. Google's real challenge isn't stretching the context window wider — it's convincing users that every extra token they stuff into it is worth the bill.
Source: ZoomBangla · TechTimes · Startup Fortune