Google DeepMind launches Gemini 3.8 Flash, beating costlier Claude Opus 5 on key benchmarks

  • Google DeepMind launched Gemini 3.8 Flash on September 2 alongside Gemini 3.8 Flash Cyber for selected government and enterprise defenders.
  • Gemini 3.8 Flash scored 54.9% on HLE-Verified and costs $0.75 per million input tokens and $3.75 per million output tokens.
  • Claude Opus 5 remains ahead on some agentic and coding tests, while Alphabet retained AdX after a federal judge rejected a forced-sale request.

Google DeepMind released Gemini 3.8 Flash on September 2, its third Flash-series model in six weeks and successor to Gemini 3.7 Flash, which launched about three weeks earlier on August 13, 2026. Priced at $0.75 per million input tokens and $3.75 per million output tokens, Gemini 3.8 Flash improves on coding, reasoning and agentic tasks and has outperformed Anthropic’s Claude Opus 5 on several benchmarks, including HLE-Verified and DeepSWE v1.1. Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens, roughly seven times more for output, but retains leads on some coding and sustained autonomous-reasoning evaluations. Alphabet also introduced Gemini 3.8 Flash Cyber for a vetted set of government and enterprise defenders through Fairwind, expanded Gemini Enterprise with pay-as-you-go pricing and token discounts of up to 20%, and won a remedies decision when U.S. District Judge Leonie Brinkema rejected the Justice Department’s bid to force a sale of the AdX ad exchange in favor of conduct-based restrictions. The twin product and legal developments followed a difficult stretch that included high-profile AI researcher departures to OpenAI and Anthropic and restructuring inside Google DeepMind, as Alphabet seeks to regain AI momentum while advertising and cloud remain strong.

The information on this website is generated using AI and we cannot guarantee its accuracy. Please use it as reference information only.