第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均;资料每比最新的一份旧 180 天,权重减半。
Mistral[5] 的公告 'Mistral Small 3.1' 日期为 2025年3月17日,其更新日志写明 'March 17: We released Mistral Small 3.1 (mistral-small-2503)';Maginative[4] 的报道日期相同。
eWeek's 18 March report notes Mistral's[1][5] own figures show it ahead of Gemma 3 and GPT-4o mini only 'in some benchmarks', and that the hardware claims apply to quantized versions.
Mistral's[1][5] model card: 24B parameters, 128k context, Apache 2.0, vision understanding added to Small 3, and the instruct scores MMLU 80.62%, MATH 69.30%, GPQA Diamond 45.96%, HumanEval 88.41%, MMMU 64.00%, DocVQA 94.08%, RULER 128K 81.20%; it notes Gemma 3 27B IT scores higher on MATH and Claude 3.5 Haiku higher on RULER 128K.
Same-day report: text and image inputs, 128K context, Apache 2.0, runs on a single RTX 4090 or a 32GB Mac when quantized, on Hugging Face[3] and Google Cloud Vertex AI with NVIDIA NIM and Azure AI Foundry to follow; gives no scores, so the claims rest on Mistral's[1][5] charts.
Mistral's[1] changelog entry 'March 17: We released Mistral Small 3.1 (mistral-small-2503)'; the same page lists Small 3.2 on June 20, 2025.
有遗漏或错误吗?用你自己的话说明:能佐证此图钉的链接、不同的开始或结束日期及理由,或缺失、有误的信息。AI 会对照此图钉的来源进行核实,搜索更好的来源,并添加任何支持你说法的页面。图钉自身的来源仍然最重要。AI 也会查看图片:显示的是别的东西或显示效果差的图片会被移到后面或替换。
原标题: Mistral Small 3.1 (24B) Released