llm-gemini plugin fix adds xhigh reasoning level for Gemini 4 Hydrogen and Neon models
Simon Willison has released version 0.38.1 of the llm-gemini plugin, which fixes the missing xhigh reasoning level for the Gemini 4 Hydrogen and Gemini 4 Neon models. The Gemini 4 Carbon model supports only low, medium, and high.
Simon Willison has released a patch version 0.38.1 of the open-source llm-gemini plugin for his command-line tool llm. The fix adds the xhigh reasoning level, which according to the author is supported by Google's Gemini 4 Hydrogen and Gemini 4 Neon models, but which was missing from the plugin since these models were added to it. According to the author, the Gemini 4 Carbon model supports only the low, medium, and high levels.
According to Willison, the xhigh level has a significant effect on both the result and the price for the Gemini 4 Hydrogen model. In his benchmark test (generating an image of a pelican on a bicycle), the model at the high level generated 5 311 output tokens for 3.99 cents, while at the xhigh level it generated 31 477 output tokens for 23.61 cents. According to the author, the result at the xhigh level was comparable to the output of the Gemini 4 Neon model.
This is a minor fix to one specific open-source tool, not a new feature of the models themselves.
Why it matters
For developers using the llm tool and the llm-gemini plugin to work with the Google Gemini API, it's relevant that without updating to version 0.38.1, it was not possible to use the highest reasoning level xhigh on the Hydrogen and Neon models, even though the models support it. The example also shows that a higher reasoning level noticeably increases output token consumption and therefore the price of a query (roughly sixfold according to the test described) — useful information when deciding whether a higher reasoning level is worth the cost for the benefit gained.
Relevant practical impact
What this means
For individuals
Anyone using the llm tool with the llm-gemini plugin and the Gemini 4 Hydrogen or Neon models needs to update the plugin in order to use the xhigh reasoning level, and should expect a significantly higher price per query at this level.
Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.