API leak reveals Gemini 3.5 Pro X-High reasoning

Google is directly answering the test-time compute escalation from competitors by adding a new top gear to Gemini's reasoning engine.

• –Gemini 3.5 Flash already introduced four thinking levels up to "high"; this leak confirms the Pro variant will push inference scaling even further.
• –The explicit focus on long-horizon tasks suggests Google is optimizing for autonomous coding agents, an area where Claude and OpenAI currently compete fiercely.
• –Exposing this via the API gives developers explicit control over the latency and cost tradeoffs for deep planning tasks.