Google Introduces Gemini 3.8 Flash and 3.8 Flash Cyber

My take: Gemini 3.8 Flash is the third Flash model Google has released in six weeks, and this time the benchmarks are concrete: with 90.8% on Terminal-Bench 2.1, it surpasses Claude Opus 5 (89.1%) and GPT-5.6 Sol (88.8%) on autonomous agent tasks. For teams building agentic workflows today, that margin matters, and it arrives at the most affordable tier in Google's ecosystem.

The release also includes Gemini 3.8 Flash Cyber, a variant restricted to Fairwind Program defenders for cybersecurity: it reaches 86.2% on the CyberGym benchmark and exceeds a 70% real-world vulnerability discovery rate, numbers that outperform much larger models.

If you are running agents today on a high-priced frontier model, these results are a signal worth acting on: is it still necessary to pay more when the "Flash" tier already outperforms on the metrics that matter most?