Gemini 4 Hydrogen: Set thinking_level, Skip Sampling Params

Google Cloud's developer guide for Gemini 4 Hydrogen lists four thinking levels, LOW, MEDIUM, HIGH (the default) and XHIGH, and says to control reasoning depth with thinking_level. It also says the backend ignores temperature, top_k and top_p, and that thinking_budget or MINIMAL returns an HTTP 400 error. Start from this request and change only the level.

from google import genai
from google.genai import types

client = genai.Client(enterprise=True, project="{{PROJECT_ID}}", location="global")

response = client.models.generate_content(
    model="gemini-4-hydrogen",
    contents="{{TASK_DESCRIPTION}}",
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(
            # HIGH is the default. Valid: LOW, MEDIUM, HIGH, XHIGH.
            # XHIGH: deep reasoning and complex multi-file engineering.
            # HIGH: multi-step agentic tool loops and code generation.
            # MEDIUM: general coding, tool use, latency-sensitive tool turns.
            # LOW: classification, routing, summarization.
            thinking_level="HIGH",
            # Uncomment if your harness edits or compacts earlier turns:
            # prefix_mismatch_behavior="DROP",
        ),
        # Do not set thinking_budget or MINIMAL (HTTP 400).
        # temperature, top_p and top_k are ignored by the backend.
    ),
)
print(response.text)

Replace the project ID and task, then pick the level that matches the workload.