A dot release to the llm-gemini plugin just landed, and it quietly unlocks a big capability for anyone running Gemini 4 Hydrogen or Neon models. Until now, the plugin only surfaced the low, medium and high reasoning levels, missing the extra xhigh setting that Hydrogen and Neon support. If you wondered why you couldn't get the kind of output you saw in some benchmarks, this is why.
The fix is in 0.38.1, and the difference is not subtle. With xhigh reasoning enabled on Hydrogen, Simon Willison reports that one prompt produced 31,477 output tokens at a cost of 23.61 cents. For comparison, the same prompt at the high level yielded just 5,311 tokens and cost 3.99 cents. That's a sixfold jump in output and price, and in this case, the output was a pelican that stands up to what Neon delivers.
What changed in 0.38.1
- The plugin now exposes the
xhighreasoning level for Gemini 4 Hydrogen and Neon. - Carbon models remain limited to low, medium and high.
- No other functional changes are mentioned in this release.
Why this matters
The xhigh setting isn't just a checkbox for quality — it materially increases the amount of output you can generate, which matters for code generation, long-form answers or any workflow where you want the model to go deep. The cost rises accordingly, so you'll want to keep an eye on token usage if you're running at scale.




