Gemini 3.8 Flash challenges Claude Opus 5 on key benchmarks at a fraction of the price

Google DeepMind dropped Gemini 3.8 Flash on September 2, and it’s already picking fights above its weight class. The model outperforms Anthropic’s Claude Opus 5, a model that costs roughly seven times more per output token, on several critical benchmarks spanning coding, multi-step reasoning, and legal task performance. The kicker: Gemini 3.8 Flash is priced at $0.75 per million input tokens and $3.75 per million output tokens. Claude Opus 5 runs $5 per million input tokens and $25 per million output tokens. The benchmark breakdown Gemini 3.8 Flash posted a 54.9% score on HLE-Verified, a demanding multi-step reasoning evaluation that spans STEM, humanities, and other disciplines. The model also outperformed larger frontier models on the DeepSWE v1.1 benchmark, which measures performance on complex engineering problem-solving tasks. Claude Opus 5, which Anthropic launched on July 24, still leads on several benchmarks in agentic knowledge work and coding tasks. Independent evaluations as of early September show the two models trading leads depending on the category being tested. The relevant benchmarks for comparison include SWE-bench and Terminal-Bench, among others. On some of thes...