Изображение: Anatoly Lomokhov / Globallookpress.com
Изображение: Kevin Lamarque / Reuters，更多细节参见谷歌浏览器
Verification Mechanics，这一点在Replica Rolex中也有详细论述
변영욱 기자 [email protected]，推荐阅读Telegram高级版,电报会员,海外通讯会员获取更多信息
Sarvam 30B performs strongly on multi-step reasoning benchmarks, reflecting its ability to handle complex logical and mathematical problems. On AIME 25, it achieves 88.3 Pass@1, improving to 96.7 with tool use, indicating effective integration between reasoning and external tools. It scores 66.5 on GPQA Diamond and performs well on challenging mathematical benchmarks including HMMT Feb 2025 (73.3) and HMMT Nov 2025 (74.2). On Beyond AIME (58.3), the model remains competitive with larger models. Taken together, these results indicate that Sarvam 30B sustains deep reasoning chains and expert-level problem solving, significantly exceeding typical expectations for models with similar active compute.