Google has launched Gemini 3.8 Flash, its third budget model in six weeks, designed to outperform its predecessor in coding tasks. The model is available in two versions: a general-purpose reasoning and coding model, and a specialized cybersecurity version called 3.8 Flash Cyber. According to Google, Gemini 3.8 Flash achieves 73.7 percent on the DeepSWE v1.1 benchmark, surpassing several competitors like GPT-5.6 Sol and Claude Sonnet 5. The model also demonstrates improved performance in 3D generation, as shown in a video of a 3D game created from a single prompt using Google's Antigravity tool. Source: thedecoder
The model is priced at $0.75 per million input tokens and $3.75 per million output tokens, matching the previous 3.7 Flash version. Google plans to increase prices to $1.50 and $7.50 starting in January 2027. Despite the lower per-token cost, the model consumes more tokens due to enhanced reasoning steps and iterative tool calls, which Google attributes to its improved performance. For efficiency-focused workloads, the company recommends using the older 3.7 Flash version. Source: thedecoder
Gemini 3.8 Flash Cyber, the cybersecurity variant, is not publicly available and is distributed through the Fairwind Program to government agencies and critical infrastructure operators. On the CyberGym benchmark, it scores 86.2 percent, outperforming the previous 3.5 Flash Cyber and several other models. The model also shows resilience against prompt injection attacks, achieving an attack success rate of 5.5 percent on the Gray Swan IPI benchmark. Source: thedecoder
