Google Prepares Gemini 3.8 Flash: Company Devs Preferred It Over Claude Opus in Coding Tests
·
02.09.2026 17:45
·
2 min read
Google is preparing to release an updated Gemini 3.8 Flash model focused on coding and autonomous agent work. According to an investigation by The Wall Street Journal, the model is expected to launch in the coming days, though no official announcement from the developers had been made at the time of writing.
For now, Gemini 3.7 Flash remains the current version on the Google DeepMind portal. Hasty online reports that version 3.8 is already fully deployed to the public are inaccurate: the service is currently in the final stage of closed testing.
Internal Tests and the Skimaki Codename
Google engineers are testing the new model’s capabilities through the internal development environment Jetski. According to WSJ sources, in a series of blind comparative tests, the company’s programmers preferred Gemini 3.8 Flash’s code generation, choosing it more often than the competing Claude Opus model from Anthropic.
As Android Authority notes, the project is known internally under the codename Skimaki. The model is being optimized for multi-step engineering tasks, long sessions with large context, and control of system interfaces. Testers also note that the new version provides more concise answers without unnecessary explanations and is less prone to factual hallucinations.
Focus on Speed and Agentic Scenarios
The Flash line was originally designed as a balance of high speed and low cost per request. In the previous version, 3.7, developers introduced a one-million-token context window and support for direct calls to system functions.
With the 3.8 release, the company aims to close the technological gap with Anthropic and OpenAI solutions specifically in applied software development. Google will publish exact API pricing and technical documentation during the official release.
