Google's New Gemini 3.8 Flash Model Shows Improved Benchmarks

Google has released its newest Flash-tier model, Gemini 3.8 Flash, on September 2, 2026. This latest addition to the Gemini family is designed for mid-difficulty long-horizon software engineering, autonomous agents, and enterprise workflows.

The Gemini 3.8 Flash retains the same introductory pricing as its predecessor, 3.7 Flash. Pricing remains at $0.75 / 1M input and $3.75 / 1M output until December 31, 2026. From January 1, 2027, prices will increase to $1.50 / $7.50.

The new model has shown improved benchmarks in several areas, including multimodal understanding (video + charts), domain-specific professional agents (finance, legal, biology), and mid-difficulty terminal/agentic coding. The Gemini 3.8 Flash excels in Vals Finance Agent v2 with a score of 61.4%, Harvey's Legal Agent Benchmark at 10.0% (best among all models), and LVBench long video with 87.8% agentic / 87.1% static scores.

However, the Gemini 3.8 Flash lags behind in certain areas, such as DeepSWE v1 (hard long-horizon SWE) at 71.0%, GDPVal-AA v2 knowledge-work Elo with a score of 1545, and Terminal-bench 4.0 (harder general agents) at 19.1%.