Gemini 4 Argon nearly matches GPT-6 Astra on the Intelligence Index

On the Intelligence Index, Gemini 4 Argon (High) scored 52.6 points, nearly matching GPT-6 Astra (Max) at 52.7; Claude Opus 5.5 (Max) scored 57.6. That composite result compares performance across one index, not every task a model might handle. Google also reported separate results on two coding benchmarks, while a small set of 3D-generation prompts offers a task-specific look at Argon’s outputs.

The Intelligence Index comparison

The three scores sit close together for Argon and GPT-6 Astra, while Claude Opus 5.5 scores higher on this index. The Intelligence Index combines 10 evaluations, so its score is a composite measure rather than a result from one task.

The difference between Argon and Astra on this index is 0.1 point. Opus 5.5’s score is higher, but the figures describe these models at the listed settings on this composite measure.

Google’s reported benchmark results

Google reported that Gemini 4 Argon scored 77.9% on DeepSWE v1.1. The company also reported 68% on CWE-bench v1, where it said Argon tied for first.

These are results on named benchmarks, separate from the Intelligence Index score. Each offers another view of performance on its particular evaluation.

What a small 3D-generation test adds

A set of 3D-generation prompts offers a practical look at visual outputs: the examples include the Golden Gate Bridge, sanctuary islands, a game, the Tower of Babel and the White House. Results vary from one prompt to another, making the sequence a task-specific complement to the composite scores.

NeoTeo has also covered questions about Argon’s practical coding performance.

Argon’s staged access

In its September 30, 2026 announcement, Google said initial access was rolling out to selected trusted cyber defenders through the Fairwind Program. The company said later access was planned for developers, enterprises and consumers, beginning with paid API customers and Google AI Ultra subscribers.