Gemini 4 Argon Benchmarks: How Google's New Model Stacks Up Against GPT-6 and Claude

Google has released the results of its latest benchmark tests for Gemini 4 Argon, showcasing how the new model performs against rivals GPT-6 Astra and Claude Opus 5.5. According to Google, Argon emerged as the top performer or tied for first place on 14 out of 19 benchmarks, covering a range of enterprise-level tasks such as professional knowledge work, agentic coding, science, long-context reasoning, and cybersecurity.

In the professional knowledge work category, Argon outperformed GPT-6 Astra by a significant margin, scoring 19.6% on the Harvey Legal Agent Benchmark, compared to Astra's 5.4%. Argon also performed exceptionally well in coding tasks, leading DeepSWE v1.1 with a score of 77.9%, compared to Opus 5.5's 74.2% and Astra's 74.1%. Moreover, Argon excelled in long-context reasoning, achieving a score of 91.9% on the Vibe Code Bench, surpassing Opus 5.5's 74.1% and Astra's 67.0%.

Independent testers have confirmed some of these findings. Vals AI ranked Argon as the top model among 41 models on its Vals Index, scoring 68.90%. Artificial Analysis also recognized Argon's strong performance, giving it a score of 53 on its Intelligence Index, which is tied with GPT-6 Astra and 23 points higher than Google's previous model, Gemini 3.1 Pro Preview.

However, the results are not a complete victory for Argon. In the FrontierSWE v2 benchmark, GPT-6 Astra achieved a higher score of 65.5% compared to Argon's 55.0%. Claude Opus 5.5 also won Terminal-Bench 4.0, a test for terminal-driven agent work, with a score of 66.4% compared to Argon's 57.4% and Astra's 68.1%. Furthermore, Astra led computer-use tasks with a score of 72.6% on the OSWorld-2.0 offline subset, while Opus 5.5 performed better in PostTrainBench for machine-learning engineering with a score of 49.3% compared to Argon's 45.3%.

It is important to note that every one of the 19 results was independently reported by Google, as is standard practice for a flagship release. Nevertheless, it is essential to remember that the model was designed with long-horizon deep work in mind, featuring a 1-million-token output ceiling, which is itself a headline feature.

In summary, while Gemini 4 Argon leads the benchmark count, independent testers still confirm its exceptional performance, and Google's rival models continue to dominate in specific areas that matter to businesses. The competitive landscape in the language model race shows no signs of slowing down, with the emergence of Argon as the new front-runner.