# llm-gemini 0.38.1 adds xhigh reasoning level for Gemini 4 Hydrogen and Neon

published: 2026-11-21T22:05:29.267Z
topic: infrastructure
source: https://simonwillison.net/2026/Nov/20/llm-gemini/
url: https://aiagentnews.io/stories/llm-gemini-0-38-1-adds-xhigh-reasoning-level-for-gemini-4-hydrogen-and-neon

Simon Willison has released llm-gemini 0.38.1, a dot release of his LLM plugin for Google's Gemini models. It adds the xhigh reasoning level to Gemini 4 Hydrogen and Gemini 4 Neon, which had been left out when those models were added to the plugin.

The llm-gemini plugin gives the LLM tool access to Google's Gemini family of models. Gemini 4 Carbon supports only low, medium and high reasoning levels, while Hydrogen and Neon also offer xhigh. Willison says he missed that difference when he added the models, and this release corrects it.

Willison's Gemini 4 comparison grid shows how much the setting changes results for Hydrogen. At high, the model used 5,311 output tokens and cost 3.99 cents. At xhigh, it used 31,477 output tokens and cost 23.61 cents, and produced a pelican drawing that he describes as competitive with the one from Gemini 4 Neon.

Willison has updated the grid with these runs, and he calls the xhigh setting a significant change for Hydrogen. The fix is limited to the plugin's model list and reasoning options.

## Why it matters if you build agents

If you call Hydrogen or Neon through llm-gemini, you can now request the xhigh reasoning level instead of stopping at high. Higher reasoning levels raise output token counts and cost, as the Hydrogen figures show, so compare cost per run before changing defaults. Carbon does not offer xhigh, so code that sets reasoning levels should check which levels each model supports.
