llm-gemini 0.38
- Source
- simonwillison.net
- Date

Context
Version 0.38 of llm-gemini, the plugin that lets Simon Willison's command-line tool call Google's Gemini models, adds Gemini 4 Neon, run as `llm -m gemini-4-neon`. Google announced Neon the same day, November 18, 2026, and Willison's release note is mostly about its price, because for the next two weeks Neon turns the Gemini 4 lineup upside down.
Google's announcement places Neon between Carbon, which it says tackles "your hardest, long-horizon projects", and Hydrogen, which "handles fast everyday execution". Its pricing page charges $3 per million input and $15 per million output tokens for Neon. Tokens are the chunks of text models are billed by, and the output rate also covers thinking tokens, the reasoning a model writes before it answers. Carbon is still on introductory rates of $2 and $10 through November 30, rising to $4 and $20 on December 1, while Hydrogen costs $1.50 and $7.50 with no change scheduled. So until December, the model Google presents as the middle option costs more per token than its flagship, and the order only matches the positioning once Carbon's price doubles. Gemini 4 Argon, which the pricing page still lists at Carbon's rates on the same schedule, is missing from the lineup Neon's announcement lays out, and Willison dropped it from his comparison grid. His note on Hydrogen's launch a week earlier had already called the Gemini 4 pricing confusing.
Per-token price is only part of what a task costs, because models differ in how many tokens they spend getting there. Google says that at its default high thinking level Neon has a 32% lower cost per task than Carbon, measured on the Artificial Analysis Intelligence Index, a third-party suite. That is Google's own figure. Willison's test is informal: his usual drawing of a pelican riding a bicycle, run at low, medium and high reasoning, which he judged better than Hydrogen's and worse than Carbon's. Using the model through the plugin also needs a paid Gemini API key, since the pricing page lists Hydrogen as the only Gemini 4 model on the free tier.
- Once Carbon doubles to $4/$20 per million tokens on December 1, the Gemini 4 lineup runs from most to least expensive as Carbon, Neon ($3/$15) and Hydrogen ($1.50/$7.50), which is the order Google's new lineup graphic shows.
- Google's Neon announcement graphic leaves out Gemini 4 Argon, which had been priced the same as Carbon, and Willison has dropped Argon from his own Gemini 4 comparison grid.
- Willison ran his pelican drawing test at Neon's low, medium and high reasoning levels, and in his Gemini 4 comparison grid Neon's pelicans look better than Hydrogen's and worse than Carbon's.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
Checking sign-in…
Loading comments…


