DeepSeek V4.1 Flash 发布，传 GPT-6.1 Astra 因安全测试未过推迟
DeepSeek 发布 V4.1-Flash-1017，官方基准称与 Claude Opus 5.5 相当、多数编码与 agent 基准超过 GPT-5.6 Sol；帖中顺带称 GPT-6.1 Astra 因内部安全测试不达标而搁置。
DeepSeek just released its most advanced model yet: DeepSeek-V4.1-Flash-1017. The company's own benchmarks put it roughly on par with Claude Opus 5.5, a significant jump over the regular V4.1 Flash. It beats GLM-5.3, and even GPT-5.6 Sol, on most coding and agent benchmarks! With GLM-5.5 and Qwen 4 also expected soon, we can safely say China is back. Expect a response from the United States, with all eyes on Anthropic and the rumored Fable 5.5. Meanwhile, no new frontier model from OpenAI. GPT-6.1 Astra is on hold after falling short in internal safety testing. Open Source vs Closed Source rematch?
这条是引用着另一条写的：x.com/deepseek_ai/status/2111375640717033777