Gemini 4 Argon Ranks Fifth on Google's Android Bench List

Google has updated Android Bench, its benchmark for measuring how well AI models help with Android coding. The refresh adds the company's own Gemini 4 Argon to the list, but the model landed well behind the latest Claude and GPT releases.

In the current ranking, Gemini 4 Argon sits in fifth place, while Claude Sonnet 5.5 holds the top spot. The results suggest that, despite being a Google product, Argon is for now trailing its rivals when it comes to Android coding assistance.

What Android Bench is for

Android Bench was introduced earlier this year to give Android developers a quick way to see which AI model is best suited to Android coding support. The test, prepared by Google, takes into account the aspects of AI use that are specific to Android coding, as well as best practices in Android development.

As Google has put new models through the test and ranked them, it has also added average cost information to the list. That means developers can look beyond raw performance and pick the option that fits their own needs.

What the October 26 update added

The October 26 update added the level of reasoning used for each model. It also brought in results for OpenAI's GPT-6 Sol and GPT-6.1 Sol, Claude Sonnet 5 and Claude Sonnet 5.5, and Google's Gemini 4 Argon.

At the top is Claude Sonnet 5.5, with a 40 percent success rate and an 80.9 percent completion rate. Claude Opus 5.5, which joined the list at the start of October, is in second place.

GPT-6.1 Sol stands out on cost

One of the more notable entries in the update is GPT-6.1 Sol. The model reached a 29.3 percent success rate and an 82.8 percent completion rate, at a listed cost of just $99.50. The top-ranked Claude model, by comparison, costs $1,574.90.

That gap shows GPT-6.1 Sol achieving a competitive completion rate at a far lower cost. Including cost in the list is meant to make precisely this kind of comparison easier.

A weak debut for Gemini 4 Argon

Gemini 4 Argon entered the list with a rather subdued showing. It is in fifth place, behind Claude Fable 5.1. Its success rate is only 21.3 percent, its completion rate is 81.8 percent, and its average cost is $267.70.

Shortly after Gemini 4 Argon was announced, a report claimed that some internal sources said the model struggled with certain coding tasks. Google, for its part, stood behind the model's capabilities.