Gemini 3.7 Flash just beat Claude Sonnet 5 and GPT-5.6 on coding benchmarks — and it's cheaper

16 August 2026

Google released Gemini 3.7 Flash on 13 August 2026, explicitly tuned for software coding and autonomous business workflows. The headline isn’t just the release — it’s the scoreboard. Gemini 3.7 Flash now beats Claude Sonnet 5 and GPT-5.6 Terra on 11 of 18 tested benchmark categories, including a leading Code Arena web-development score of 1,588 Elo. It’s priced at 75 cents per million input tokens and $3.75 per million output tokens through the end of the year, undercutting both rivals on cost while leading on several coding measures.

This is the fourth major reordering of the “best AI coding model” leaderboard this year, following Claude Opus 5’s launch, GPT-5.6’s rollout, and Cursor and Windsurf both re-pricing around whichever model was fastest that month. None of these labs are winning by a wide, stable margin anymore — they’re trading the lead every few weeks, each time on a different axis (raw reasoning, coding accuracy, speed, or price). For a market that got used to one obvious “best” model, that’s now the wrong question to be asking.

So what

If your business is choosing an AI coding stack — or a development partner’s stack — based on which model was best when you last checked, that decision has a shelf life measured in weeks, not years. The practical response isn’t to chase the leaderboard; it’s to work with a build process that isn’t locked to one vendor’s model and can move to whichever tool is genuinely fastest and most reliable for the job at hand. That’s how we run AI-assisted builds — see our AI-assisted development approach, or get in touch if you want a build that isn’t hostage to one lab’s next price change.

← All signals