Claude or Gemini - which is better for coding?

Neither wins outright: Gemini 4 Argon leads on DeepSWE, while Claude Opus 5.5 is ahead on FrontierSWE and Terminal-Bench, according to Google's own model card. The better choice depends on the kind of coding work you do and the tools you already use.

• Benchmarks: On Google's model card, Gemini 4 Argon scores 77.9 on DeepSWE v1.1 versus 74.2 for Claude Opus 5.5. Opus 5.5 leads on FrontierSWE v2 (62.3 vs 55.0) and Terminal-Bench 4.0 (66.4 vs 57.4).
• A cheaper Claude option: Anthropic reports Claude Sonnet 5.5 at 70.6 percent on Terminal-Bench 4.0. Vendor figures from different sources and test harnesses are not directly comparable.
• Price: Gemini 4 Argon costs 2 / 10 USD per 1 million input / output tokens until 30 November 2026. From 1 December 2026 it rises to 4 / 20 USD, the same as Claude Opus 5.5. Claude Sonnet 5.5 costs 2 / 10 USD, and Claude Haiku 5.5 costs 0.10 / 0.50 USD. Argon is not on the free tier.
• Tools: Claude Code works in the terminal, VS Code, JetBrains, a desktop app and the web. Google's coding agent environment is Antigravity, where Gemini 4 Argon is available on Google AI Pro, Ultra and Enterprise plans.

Pick a real task from your own repository and run it through both Claude Code and Antigravity before you commit. If terminal-heavy agent work matters most, start with Claude. If you want to test Argon cheaply, do it before the December price change.