01 What happened
On November 12, 2026, Google Cloud released Gemini 4 Hydrogen, a short-term generally available model with a minimum 45-day retirement notice policy. Key specifications include a unified 1,048,576-token limit for both input and output, long decodes up to 1,048,576 tokens on the Interactions API and GenerateContent (v1beta1), four thinking levels (LOW, MEDIUM, HIGH by default, XHIGH), and automatic implicit caching for prompt prefixes of 4,096 tokens or more. The model accepts Microsoft Office and OpenDocument files. Sampling parameters temperature, top_p, and top_k are ignored; frequency_penalty, presence_penalty, and candidate_count greater than 1 return HTTP 400 errors.
02 Key details
- Gemini 4 Hydrogen is a short-term GA release subject to a minimum 45-day retirement notice policy.
- The model has a unified 1,048,576-token input and output limit; the default single-segment foreground output cap is 65,536 tokens.
- Four thinking levels are available: LOW, MEDIUM, HIGH (default), and XHIGH; thinking_budget and MINIMAL return HTTP 400 validation errors.
- Migration from Gemini 3.8 Flash requires removing temperature, top_p, top_k, frequency_penalty, presence_penalty, and candidate_count > 1; implicit caching for prefixes of 4,096 tokens or more is enabled by default.
03 Why it matters
The model provides a significantly larger context window and new capabilities for agentic workflows, but requires migration from previous versions, including SDK upgrades and removal of deprecated parameters.
04 Who it matters to
Software developers and enterprise AI customers.