llm-gemini 0.35

Simon Willison has published llm-gemini 0.35, a new version of his plugin that connects the LLM command-line tool to Google's Gemini models. The release adds support for Gemini 4 Argon, a model Google made available today through the Gemini API and to customers on its $100-per-month subscription plan.
Willison tested the model with his usual pelican-on-a-bicycle SVG prompt at low, medium and high effort settings, and reports that all three outputs were excellent. He notes that Google has a strong track record with SVG generation and is the only lab he knows of where staff have invested in SVG work specifically, including past pelican demonstrations.
The high-effort illustration cost 17.166 cents to generate. That figure reflects Argon's launch pricing: $2 per million input tokens and $10 per million output tokens.
Those rates put Argon at the same price as GPT-6 Sol, half the price of Claude Opus 5.5, and roughly one-fifth the price of both Claude Fable 5.1 and GPT-6 Astra. The pricing is scheduled to change on December 1st, when Argon is set to double in cost, which would bring it level with Claude Opus 5.5.
Willison argues that sticker prices are less meaningful than they once were, because what actually matters is how many tokens a model consumes to complete a goal inside an agentic harness. A useful signal on that front comes from Artificial Analysis, which publishes the total cost of running its benchmark with each model. Argon comes in at $2,407 for that benchmark, placing it close to GLM-5.3 and well below the latest models from OpenAI, Anthropic and Meta. Even after the December price increase, Willison expects Argon to remain somewhat cheaper than GPT-6 Astra on that measure.
For developers, the practical effect of this release is that Argon can be called through the LLM tooling without writing custom API code. The LLM project is an ecosystem of command-line utilities and Python libraries for working with language models, and plugins like llm-gemini extend it to individual providers. Anyone already using the plugin can pick up the new model by upgrading the package rather than integrating Google's API directly.
This suggests the more interesting number here is not the per-token rate but the benchmark cost. Token efficiency varies enough between models that a cheaper headline price can still lose an agentic task on total spend, since a model that reasons longer or retries more burns more tokens to reach the same outcome. Artificial Analysis's aggregate figure is one attempt to capture that, though it reflects a single benchmark and should be treated as a directional signal rather than a guarantee for any particular workload. Still, Argon landing near GLM-5.3 while sitting below the flagship offerings from several major labs is a notable position for a model launched today, and the scheduled doubling means anyone evaluating it for production should model costs against the post-December rate rather than the introductory one.
Willison also shared bonus Argon outputs, including an SVG armadillo in fishnet tights jaywalking on Mars, a Fabergé egg generated with Blender, and a video from DeepMind staffer @fofrAI of a pelican riding a bike built from Fourier epicycles. He acknowledged using ChatGPT to edit a chart accompanying his post, adding that the underlying data was verified and that the edit was needed to keep a Gemini 4 Argon info box from overlapping other numbers.
Why it matters: teams already using the LLM CLI or building on Gemini can add Argon by upgrading llm-gemini, and the price trajectory gives them a deadline to weigh. Because Argon is set to double on December 1st, any cost comparison or budget model based on today's rates will be wrong within weeks, and in my inference the benchmark cost figure is the more durable input for agent workloads than the headline per-token price.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Release: llm-gemini 0.35 I added support to llm-gemini for the new Gemini 4 Argon model, made available today via the Gemini API and for customers of their $100/month subscription plan. Here are the low, medium and high pelicans . As expected, they are all excellent: Gemini has a track record of producing great SVGs, and are the only lab where I know their staff have invested in SVG specifically (and who have boasted about their pelicans before.) Here's the "high" illustration, which cost me 17.166 cents: The prici