LLM Gemini 0.35 has been released.
llm-gemini 0.35 supports Gemini 4 Argon, highlighting launch pricing, December doubling, pelican vector tests, and agentic benchmark token costs.
Key Points Summary
- llm-gemini 0.35 adds support for Gemini 4 Argon, making the new model available through the Gemini API and the $100/month plan.
- Gemini 4 Argon has strong vector graphics output, highlighted by pelican examples and a high-quality illustration generated for about 17 cents.
- Launch pricing is $2 per million input tokens and $10 per million output tokens, but prices are set to double on December 1.
- Benchmark cost is a more important pricing signal than headline rates, with Gemini 4 Argon costing $2,407 to run Artificial Analysis’s benchmark.
- Users should reassess workflow choices before December 1, since total token usage may matter more than per-token price.
Meta description: llm-gemini 0.35 now supports Gemini 4 Argon through the Gemini API and a $100/month plan, with launch pricing, benchmark costs, and upcoming changes.
Tags: Gemini 4 Argon, llm-gemini, AI pricing, vector graphics, benchmark
Featured image: abstract friendly illustration style showing a pelican, smooth curves, and a pricing graph, no real people, no logos, no text
What the release changes
The llm-gemini 0.35 release adds support for Gemini 4 Argon, a new Gemini model made available today through the Gemini API and to customers of the $100/month subscription plan. For users of llm-gemini, this makes the newest Gemini option available in that tool. The release notes also highlight the model's output quality, its launch pricing, and a benchmark cost figure that may matter more than sticker price.
The pelican test and vector graphics
The notes include low, medium and high pelican examples, and describe all three as excellent. The reason is not just a joke: Gemini has a track record of producing great vector graphics, and the release notes say it is the only lab known to the author where staff have invested in vector graphics specifically and have previously boasted about pelican examples.
The high illustration was singled out as a concrete example. It cost 17.166 cents to generate. That small amount is part of the story because the launch pricing for Gemini 4 Argon is low enough that a single visual test can be tracked to a fraction of a cent.
Launch pricing and the December price change
At launch, Gemini 4 Argon is priced at $2/million input and $10/million output. The notes compare that to other models: it is the same price as GPT-6 Sol, half the price of Claude Opus 5.5, and 1/5th the price of Claude Fable 5.1 and GPT-6 Astra.
That launch window is short. On December 1st, the price is set to double, which will put Gemini 4 Argon at the same price as Claude Opus 5.5. The release notes say that even after the increase, it will still be a little cheaper than GPT-6 Astra.
For readers, the practical takeaway is that the model is currently in a temporary price position. A workflow that is cheap enough today may become less attractive after December 1st, especially if a user has been selecting models mainly by headline input and output rates.
Why token count matters more than sticker price
The release notes caution that prices like this are not really comparable these days. What matters is how many tokens a model takes to achieve a goal within an agentic harness, meaning a setup where the model may need several steps to finish a task. A lower per-token price can still end up expensive if the model needs many more tokens to get the job done.
A useful clue comes from Artificial Analysis, which publishes the total cost to run its benchmark using each model. Gemini 4 Argon is costing $2,407 to run that benchmark. The notes say this places it close to GLM-5.3 and significantly below the latest models from OpenAI, Anthropic, and Meta.
The author also notes that ChatGPT was used to edit the chart showing that comparison. The data was checked, and the edit was only meant to move the Gemini 4 Argon information box so it did not overlap other numbers. That detail matters because the chart is part of the evidence behind the pricing discussion, and the author is making clear that the layout change did not alter the figures.
Bonus examples and what comes next
The release notes end with a few bonus examples that show where the model is being tested. One is a prompt to generate a vector graphics image of an armadillo in fishnet tights jaywalking on Mars. Another is a Fabergé egg created using Blender. The notes also mention a video shared by DeepMind staffer @fofrAI, showing a pelican riding a bike using fourier epicycles, described as a fun new twist.
The next thing to watch is December 1st, when the price doubles. Until then, Gemini 4 Argon has a pricing position that is unusually easy to compare against other models, but the Artificial Analysis total benchmark cost is the stronger signal for workloads where token use, not headline price, drives the bill.
Comments (0)