Google plans to release Gemini 3.8 Flash, a new model with strengthened coding capabilities, as early as Wednesday, according to a Wall Street Journal report by Erin Woo1. The model, developed under the internal codename "Skimaki," has narrowed the gap in core programming capabilities between Google and the flagship models of Anthropic and OpenAI, based on internal testing3.
In internal comparison tests using Google's Jet Ski programming tool, engineers showed a stronger preference for Gemini 3.8 Flash than for Anthropic's Opus model. The result marks a concrete step for a company that has faced setbacks in model competition since the release of Gemini 3.0.
Separately, the report indicates that Gemini 4 has performed well on pre-training evaluations but still needs to complete post-training before release.
The planned launch arrives amid leadership changes at Google DeepMind. DeepMind co-founder Demis Hassabis stepped down from daily management duties last month, taking on the roles of department chairman and Alphabet chief scientist. His long-time deputy Koray Kavukcuoglu has fully assumed operational leadership and was promoted to senior vice president. Since taking over, Kavukcuoglu has required the team to accelerate execution speed.
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on September 1, pairing benchmark gains in coding and reasoning with a 75% reduction in cache-read costs ctx. ANALYSIS The timing puts Gemini 3.8 Flash in direct proximity to that Anthropic release, while Google's internal benchmarking referenced Anthropic's Opus model rather than the newer Fable and Mythos lines.
The leadership transition at Google DeepMind, with Kavukcuoglu's explicit push for faster execution, provides organizational context for the rapid cadence: shipping a coding-focused model while Gemini 4 remains in post-training suggests a strategy of incremental releases rather than waiting for a single flagship drop.