Google ships third Flash model in six weeks: Gemini 3.8 Flash tops long-horizon coding benchmarks
CodeByPoonam · x · 2026-09-03
Google released Gemini 3.8 Flash and Gemini 3.8 Flash Cyber today, its third Flash update in six weeks, both built on the same core intelligence.
3.8 Flash (general)
- Outperforms most larger frontier models on long-horizon software engineering (DeepSWE v1.1)
- Leads finance and legal reasoning benchmarks (Vals Finance Agent V2, Harvey's Legal Agent Benchmark)
- Scores 54.9% on HLE-Verified, spanning STEM, humanities, and professional fields
- Priced same as 3.7 Flash: $0.75/M input, $3.75/M output tokens
3.8 Flash Cyber
- Frontier-level vulnerability discovery, beats larger models on CyberGym with a 70%+ success rate
- Restricted to trusted defenders via Google's new Fairwind Program
Logan Kilpatrick confirmed the release, highlighting gains in agentic and coding capabilities.
Related event: Google Launches Gemini 3.8 Flash and Cyber Security Variant(104 posts)→
More from coding & agent
- Two Attribution Disputes Later: Keeping Open Source Research Viable in the Age of Coding Agents — dhruv2038 · 2026-09-03
- Meta's Muse Spark 1.3 is now free on OpenCode — ramagetime · 2026-09-03
- GitHub Announces First-Ever Copilot Day on September 10 With Product Announcements — PaulShellDev · 2026-09-03
- Inside the $950 Maven Course on Production-Ready Enterprise AI Agents — hugobowne · 2026-09-03
- Deploying an AI Agent on Modal in Six Minutes: A Walkthrough — hugobowne · 2026-09-03
- Dev Offers Scale AI 300GB of Agent Session Logs, Runs 63 Agent Subscriptions at $13.5K/Month — doodlestein · 2026-09-03