Google Launches Gemini 3.5 Pro in Enterprise Preview: 2M-Token Context and Deep Think Reasoning

Google opens Gemini 3.5 Pro in limited Vertex AI enterprise preview today: a 2.1M-token context window and Deep Think reasoning mode, rebuilt from scratch after the original model failed recursive tool-calling and SVG generation tests.

Gemini 3.5 Pro was rebuilt from scratch after Google's engineers found structural failures in the original version. The launch today is limited to enterprise Vertex AI customers who request early access — no public API listing on OpenRouter or the standard Gemini API yet.

What's reportedly shipping:

• 2.1M token context — enough for ~1,500 pages of technical documents in a single call; changes the chunking calculus for long-document analysis and multi-file review
• Deep Think reasoning — Google's extended-thinking mode, similar to Claude's extended thinking or OpenAI's o-series; gated to the Ultra tier on the consumer side, API access TBD
• Reported pricing — ~$15 input / $60 output per million tokens (vs Gemini 3.1 Pro at $2/$12), from third-party sources — not officially confirmed by Google

Gemini 3.5 Flash, the faster/cheaper sibling, is already available on the public Gemini API and OpenRouter.

Why it matters: 2M context removes the retrieval bottleneck for a class of document-analysis and code-review tasks where chunking introduces errors. Watch the Gemini API changelog and ai.google.dev pricing page for GA availability and confirmed pricing before building against it.

Sources: Gemini 3.5 Pro launch overview — EnterpriseDNA · Gemini 3.5 Pro API access guide — byteiota · Gemini API pricing — ai.google.dev