ChatGPT
Google’s latest flagship, Gemini 4 Argon, is a leader on several benchmarks—but not an across-the-board winner. Its published results are particularly strong in business automation, long-context reasoning, and video understanding; GPT-6 Astra and Claude win some demanding coding and scientific-workflow tests. ()
As of October 14, 2026, Argon is publicly available: Google announced it on September 30 and opened access in the Gemini app and developer API on October 13. ()
Key benchmark scores
For “ChatGPT,” the comparison below uses GPT-6 Astra, OpenAI’s highest-capability model—not every model available inside ChatGPT. Claude is represented by Opus 5.5, and Grok by Grok 4.7. These are published results compiled from different evaluation setups, not a single controlled four-way test. ()
Higher is better. Bold marks the highest listed score, not necessarily a statistically significant lead.
Gemini, GPT, and Opus figures come from Google’s published comparison; Grok figures come from its release report. Grok’s DeepSWE result uses high effort; its other listed results use xHigh. “—” means absent from these source tables, not zero capability. ()
What the results mean
- Gemini: The clearest advantages here are automation and long-context reasoning. Its coding results are mixed: it leads DeepSWE but trails on FrontierSWE and terminal tasks. ()
- ChatGPT / GPT-6 Astra: Stronger on the listed scientific workflows, FrontierSWE, and OSWorld computer-use test. The newer GPT-6.1 Sol is a lower-cost alternative, so Astra is not a proxy for every ChatGPT experience. ()
- Claude: Opus leads terminal tasks in this table. Importantly, Sonnet 5.5 separately reports 70.6% on Terminal-Bench 4.0, illustrating why choosing one model per company can obscure stronger results elsewhere in its lineup. ()
- Grok: Grok 4.7 matches Argon’s published legal-workflow score, but has lower listed DeepSWE and terminal-task scores. Missing results prevent a defensible overall ranking across all four. ()
Important caveats
Google generally uses maximum reasoning settings and draws competitor scores from provider reports or benchmark leaderboards. Agent software, tool access, budgets, and safeguards differ; its video evaluation also supplies different numbers of frames to different models because of API limits. These are useful comparisons, but not perfectly apples-to-apples. ()
Bottom line: Argon looks especially competitive for large-document and multimodal workflows. For coding and scientific work, there is no universal winner—the benchmark and exact model matter more than the company name. ()