AivexaNewsSearch
AI news for builders and product teamsChecked every hour

llm-gemini 0.36

Collected Nov 10, 2026

Simon Willison published version 0.36 of llm-gemini, his plugin that connects Google's Gemini models to the LLM command-line tool and Python library. The release adds support for Gemini 4 Carbon. According to the post, Google had announced Carbon two weeks earlier for "a subset of cyber defenders," and as of today it is available to everyone else. Willison calls it Google's best Gemini so far.

Pricing is the detail most teams will weigh. Carbon launches at $2 per million input tokens and $10 per million output tokens, which Willison places at the same level as GPT-6.1 Sol and Claude Sonnet 5.5. That rate is scheduled to double on December 1st to $4 and $20 respectively, matching Opus 5.5. In practice, any product built on Carbon now should model unit economics against the December price rather than the launch price, since a doubling of inference cost can erase thin margins on high-volume workloads. That is my reading of the stated numbers, not something the post argues.

Willison also notes the model defaults to medium thinking effort. The post shows the plugin producing three pelican-on-a-bicycle drawings at low, medium and high effort, and he describes the high-effort output as his example. He says Gemini's pelicans have been strong for a while, so he moved on to a harder test.

That test used Google's Antigravity IDE with a prompt he had first tried against Claude Fable 5.2 a day earlier: a Three.js animation in which Yonkers stands up as a giant mech, with a pronounced standing-up motion, then stomps around Manhattan, Queens and Brooklyn, using OpenStreetMap mapping data. He says the result required one follow-up prompt, asking to build it into a standalone HTML file, to produce a version with no external dependencies, and he screen-captured the output as a video.

A later update in the post points to a YouTube demonstration, also shared by Google CEO Sundar Pichai, in which Antigravity's /swarm feature was used with Carbon to build an open-world 3D game. The player explores Google's Mountain View campus as a pelican riding a GBike. Willison describes the build as coordinating subagents to model 3D assets in Blender from Omni and Nano Banana references, construct the browser game in Three.js with WebGL shaders, and pair procedural Web Audio sound effects with Lyria music. A playable version is linked in AI Studio.

For developers, three things stand out. First, the plugin update means Carbon can be selected through the same LLM interface as earlier Gemini models, so switching models is a configuration change rather than a rewrite. Second, the effort setting is a cost lever: higher thinking effort generally means more tokens generated before the answer, and with output priced at five times input, that spend adds up quickly. Users should check whether the plugin exposes effort as an option so they can pin it per task instead of relying on the default. Third, the multi-tool demonstrations suggest Carbon is being positioned for agentic work that spans asset generation, code and audio, though a single showcase is not a benchmark.

Why it matters: teams already using llm-gemini can adopt Carbon today through a version bump, but they should budget for the December price change before shipping anything that depends on it. The Antigravity examples, if representative, point to more reliable long multi-step builds, which affects how product teams scope agent features. Treat both the price and the capability claims as things to verify on your own workloads.

Read at Simon Willison

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Release: llm-gemini 0.36 Gemini 4 Carbon was announced two weeks ago for "a subset of cyber defenders", but today it's available to everyone else . This is Google's best ever Gemini, priced at $2/million input and $10/million output at launch (that's GPT-6.1 Sol / Claude Sonnet 5.5 prices) but scheduled to double in price to $4/$20 on December 1st (same as Opus 5.5). Here are three pelicans for the low, medium and high thinking efforts (it defaults to medium). This is the pelican I got for high: Gemini pelicans hav