Third-party benchmarks for Gemini 4 Argon are in. The verdict: it’s specialized, but for good reason.
According to the accounting benchmark published by DualEntry, Argon scored 91.1% for accuracy, ranking first among 18 models. Its consistency score of 86.1% also ranked first. It costs $0.11 to complete a task, compared with $1.18 for Claude Fable—10 times more expensive, yet with a lower score.
But according to Artificial Analysis’s coding index, Argon’s DeepSWE score of 78.8% came in first, while its Terminal-Bench score was only 56.1%, putting it mid-pack and well behind Claude Sonnet 5.5 (66.2%).
My take: Argon is the strongest for “long-running + specialized” tasks—think finance, law, and accounting, where tasks can take hours and mistakes aren’t an option. For quick, straightforward coding, don’t blindly trust it.
Look at the whole report, not just the parts you like.
DYOR, and trade rationally.
According to the accounting benchmark published by DualEntry, Argon scored 91.1% for accuracy, ranking first among 18 models. Its consistency score of 86.1% also ranked first. It costs $0.11 to complete a task, compared with $1.18 for Claude Fable—10 times more expensive, yet with a lower score.
But according to Artificial Analysis’s coding index, Argon’s DeepSWE score of 78.8% came in first, while its Terminal-Bench score was only 56.1%, putting it mid-pack and well behind Claude Sonnet 5.5 (66.2%).
My take: Argon is the strongest for “long-running + specialized” tasks—think finance, law, and accounting, where tasks can take hours and mistakes aren’t an option. For quick, straightforward coding, don’t blindly trust it.
Look at the whole report, not just the parts you like.
DYOR, and trade rationally.