Google’s latest Flash model targets software engineering, autonomous agents and complex enterprise workflows at a lower cost.
Google has released Gemini 3.8 Flash, its latest AI model focused on coding, agentic tasks and complex workflows. The model was introduced on September 2 and is now generally available through Google’s AI developer services.
Google says Gemini 3.8 Flash is its most intelligent Flash model yet. It is designed for long-horizon software engineering, autonomous agents and complex enterprise workflows. The company says the model offers stronger reasoning while keeping the speed and cost advantages of the Flash series.
The launch comes only three weeks after Gemini 3.7 Flash. Google said Gemini 3.8 is its third Flash release in six weeks. The company has also introduced a separate Gemini 3.8 Flash Cyber model for cybersecurity use cases.
Built for coding and AI agents
Gemini 3.8 Flash is aimed at tasks that require several steps and tool calls. Google says it can handle software engineering, knowledge work and autonomous agent workflows. It also supports text, images, audio, video and PDF inputs.
The model has an input context limit of 1,048,576 tokens and a maximum output limit of 65,536 tokens. A context window allows an AI model to process a large amount of information in one request.
Google is offering Gemini 3.8 Flash at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens. The company says this pricing will continue through December 31, 2026. Standard pricing will rise to $1.50 per million input tokens and $7.50 per million output tokens from January 1, 2027.
Gemini 3.8 Flash performance
Google’s published evaluations show strong results in software engineering and agent tasks. On the DeepSWE v1.1 benchmark, the company says Gemini 3.8 Flash scores above 70 percent while maintaining a low cost per task.
On the Vals Finance Agent v2 benchmark, Gemini 3.8 Flash scored 61.4 percent. Claude Opus 5 scored 58.6 percent, while Gemini 3.7 Flash scored 59.0 percent. Google also reports a 10.0 percent result for Gemini 3.8 Flash on Harvey’s Legal Agent Benchmark.
The company says the model is built to improve reliability during long-running agent tasks. It is designed to navigate roadblocks and resolve coding and real-world problems with greater accuracy.
Internal testing and coding focus
The Wall Street Journal previously reported that Google internally tested the model under the codename “Skimaki”. According to the report, Google engineers compared it with Anthropic’s Opus model using the company’s internal Jetski coding platform.
However, Google’s public announcement is the stronger source for the model’s capabilities and published benchmark results. The company does not describe Gemini 3.8 Flash as simply matching one rival model across all tasks. Instead, its published data shows different results across specific benchmarks.
Availability for developers
The new model is also being used across Google’s developer tools. Gemini 3.8 Flash is the default agent for managed agents in Google Antigravity, while the Antigravity SDK also uses it by default.
For developers in India, the release could be relevant for software development and other technical workflows. Its Flash-tier pricing is designed for applications that need to run AI tasks at scale.
Gemini 3.8 Flash is now available as a stable model through Google’s AI services. Developers can access it through Google AI Studio and the Gemini API.
For the latest updates, follow CBY News
1 thought on “Google launches Gemini 3.8 Flash with stronger coding and AI agent skills”