Gemini 3.7 Flash Takes Top Spot on Coding Benchmarks at 340 Tokens/Sec

In one line: Google launched Gemini 3.7 Flash on August 13, positioning it as the fastest and highest-scoring model on production-code benchmarks — just three weeks after its predecessor.

Key points

• Benchmark lead: 43.6% on FrontierCode 1.1 Main, ahead of Claude Sonnet 5 (42.7%) and GPT-5.6 Terra (41.3%)
• Coding leap: DeepSWE v1.1 jumped from 49.0% (3.6 Flash) to 65.3% — a 16-point gain in 23 days
• Speed: ~340 output tokens/sec in independent Artificial Analysis testing, top tier among reasoning-class models
• Pricing: $0.75 / $3.75 per million tokens (introductory; rises to $1.50 / $7.50 after December 31, 2026)
• Context: 1M-token window unchanged from the previous version

Why it matters

Three weeks between Flash releases signals Google is on an aggressive release cadence in the coding-agent race. Pairing benchmark leadership with competitive speed and pricing makes Gemini 3.7 Flash a compelling default for agentic pipelines — though the introductory price expiry is worth factoring into long-term cost planning.

Read more

• Gemini 3.7 Flash Model Card — Google DeepMind
• Gemini 3.7 Flash: Pricing, Benchmarks and What Changed — FelloAI