Google's Gemini 4 Argon Takes Most Benchmark Wins, With Caveats

Google has unveiled Gemini 4 Argon, and its own comparison table places the model first on most of the benchmarks it published. By Google's count, Argon leads outright on 12 of the 18 tests and ties for first on one more. Some outlets tally it as 13 of 18, so the exact number depends on how ties and table rows are counted.

Google highlights a few results. Argon scored 77.9 percent on DeepSWE v1.1 and 91.7 percent on LVBench, and on AutomationBench it posted 51.3 percent, which Google says is the top ranking. Independent tracking from Vals also has Argon at the top of its overall index, with a score of 68.90 percent across 41 models.

The picture is less one-sided than the headline suggests. OpenAI's GPT-6 Astra leads outright on three of the disclosed benchmarks and ties Argon on another, and Anthropic's Claude Opus 5.5 leads on two. Argon also finishes last of four on FrontierSWE v2 and Terminal-Bench 4.0 in Google's table. Another analysis reports Claude Opus 5.5 ahead on Terminal-Bench 4.0 by 66.4 percent to 57.4 percent. Those gaps matter for anyone choosing a model for coding agents or terminal-based work.

Access is the other limit. Argon is rolling out first to a group of trusted cyber defenders through Google's Fairwind Program, while the company works through a voluntary U.S. government pre-release review process. Most developers and businesses cannot test it yet, so the claims cannot be checked against everyday workloads.

For now, the safest reading is that Gemini 4 Argon is a strong new contender with a broad lead on the tests Google chose to publish. Whether that lead holds up in wider release, and on tasks outside the vendor's selection, is the question to watch.