Gemini 4 Argon scores 77.9% on DeepSWE, ahead of Opus 5.5 and GPT-6 Astra
Gemini 4 Argon reaches 77.9% on DeepSWE versus Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%, but still trails Opus on Terminal-bench and Astra on FrontierSWE.
Gemini 4 Argon is out, and the DeepSWE number looks strong: 77.9% vs Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%. But it doesn't win everywhere. Opus still leads Terminal-bench, and Astra leads FrontierSWE. Pricing is $2 / $10 per 1M tokens at launch, then $4 / $20. Same as Opus. Right now it's only for selected cyber security teams. Paid API and AI Ultra come next, no date yet. I'll take these scores with a grain of salt until someone reproduces them. Gemini 3 tool calling broke a lot in agent workflows, so I want to test it myself first.
This post quotes another post: x.com/4783690002/status/2105388084154056939