Google released Gemini 3.8 Flash on September 2, 2026 -- internally known during development as "Skimaki" -- as a concentrated push to close the gap on agentic coding specifically, the capability increasingly cited across the industry as generative AI's top enterprise use case, and the one metric where Google has trailed OpenAI and Anthropic most consistently.
In head-to-head internal testing on Google's own Jetski developer platform, engineers reportedly preferred the new model over Anthropic's flagship Claude Opus -- a notable internal result given how consistently third-party public benchmarks have favored Claude and GPT models specifically on coding tasks through 2025 and into 2026.
Worth treating an internal preference test with the caution it deserves until independent public benchmarks -- SWE-bench Verified, LiveCodeBench, and similar -- confirm it. But the fact that Google is racing to close this one specific gap, rather than pursuing a broader general-capability push, is itself informative about where the real competitive battle line currently sits among the frontier labs.
Gemini 3.8 Flash's internal coding-preference result over Claude Opus is worth watching for independent confirmation, not taking at face value -- but Google concentrating specifically on the coding gap, rather than general capability, is a clear signal of where it sees itself behind.