llm-gemini 0.38.1

Simon Willison has patched a configuration error in his llm-gemini plugin to correctly expose the xhigh reasoning level for Gemini 4 Hydrogen…

By Vane November 21, 2026 1 min read
llm-gemini 0.38.1

Simon Willison has patched a configuration error in his llm-gemini plugin to correctly expose the xhigh reasoning level for Gemini 4 Hydrogen and Gemini 4 Neon. The original build mistakenly limited all three models to low, medium, and high settings, omitting this extra tier available only on the latter two variants. This update ensures developers can now access the specific instruction set required to trigger deeper processing chains.

The practical impact is visible in the quality and cost of generated output. A comparison of a pelican riding a bicycle shows that the xhigh setting produces a more detailed image than the high setting, matching the quality of the more expensive Neon model. While the high setting consumed 5,311 output tokens costing 3.99 cents, the xhigh setting used 31,477 output tokens costing 23.61 cents to achieve a superior result. This demonstrates that the additional reasoning steps significantly increase token usage and expense while improving visual fidelity.

* Hydrogen xhigh output tokens: 31,477
* Hydrogen xhigh cost: 23.61 cents
* Hydrogen xhigh time: 2m 36s

Scroll to Top