El diario
Retratos de modelos
DeepSeek V4 Flash 与 GPT-6.1 Sol 摘要对比实操
从版本核验、样本准备到盲评和成本核算，逐步比较两款模型的摘要表现，避免把接口别名、文字流畅度或宣传指标误当成可靠结论。
Qwen3.7 Plus 大规模批处理：从试跑到可靠交付
分步搭建 Qwen3.7 Plus 批处理流程，覆盖接口核验、数据分片、思考预算、任务提交、失败重试与成本对账，并区分官方批量接口和中转服务能力。
用 Haiku 级小模型搭建可靠分类流水线
从标签定义、评测集和结构化输出，到异常分流、成本核算与灰度上线，逐步搭建可验证、可回滚的小模型分类流水线。
哪个模型最擅长处理脏乱电子表格：六步实测指南
从合并表头、混合日期到跨表公式，逐步建立可复现的测试流程，用正确率、结构保留和实际成本选择适合处理脏乱电子表格的模型。
Gemini 4 Argon 与 Claude Opus 5：分步完成工程选型
从输入能力、代码修复、工具调用到缓存账单，逐步比较 Gemini 4 Argon 与 Claude Opus 5，用可复现测试确定适合项目的模型。
逐步算清：何时值得为Opus级模型的额外质量付费
从验收标准、对照评测到缓存账单与升级策略，逐步计算Opus级模型能否通过减少返工、重试和人工复核收回额外成本。
Claude Sonnet 5 与 GPT-6 Astra：结构化输出对比实操
从统一数据契约、配置原生接口到异常处理与成本核算，逐步搭建可复现的结构化输出对比流程，以业务正确率而非格式观感选择模型。
客服机器人该选哪个模型：从真实工单到上线验收
按任务拆分、模型初筛、真实工单评测、工具验证、成本核算与灰度上线六个步骤，为客服机器人选择合适的模型。
代码生成与代码审查如何选模型：六步建立可靠流程
从任务拆分、真实仓库评测到缺陷验证和成本核算，逐步为代码生成与代码审查选择合适模型，避免只看榜单或单次报价。
Building Small, Quick Assistants on MiniMax M3, One Step at a Time
Build a focused MiniMax M3 assistant with bounded context, controlled reasoning, safe tools, and measurable latency and cost.
Qwen3.8 Max vs Closed Models: A Practical Walkthrough
Compare Qwen3.8 Max with frontier closed models through benchmark checks, controlled coding tests, latency measurements, and cost analysis.
Grok 4.7 for Current Events: A Step-by-Step Guide
Build a repeatable Grok 4.7 workflow for current events with explicit dates, verified sources, careful citations, and controlled API usage.
Gemini 3.8 Flash When Speed Matters More Than Depth
A practical walkthrough for configuring, testing, and deploying Gemini 3.8 Flash when responsiveness matters more than exhaustive reasoning.
DeepSeek V4 Pro: A Mathematical Reasoning Walkthrough
A practical walkthrough for solving mathematics with DeepSeek V4 Pro, configuring reasoning, checking answers, and tracking API usage.
Gemini 3.1 Pro for Document-Heavy Work
Build a repeatable Gemini 3.1 Pro workflow for document intake, evidence extraction, cross-document analysis, validation, and cost tracking.
GPT-5.6 Sol vs GPT-6.1 Sol: Everyday Prompts
A practical walkthrough for comparing GPT-5.6 Sol and GPT-6.1 Sol on writing, summaries, planning, and code without inventing benchmark results.
GPT-6.1 Sol in Practice: Strengths and Limits
A step-by-step guide to evaluating GPT-6.1 Sol for coding, tool use, reliability, latency, and real production costs.
What Claude Fable 5.1 Is Good At: A Walkthrough
Explore Claude Fable 5.1 through practical steps for coding, research, document work, tool integration, and quality evaluation.
GPT-6 Astra: A Careful Reasoning Walkthrough
A practical, step-by-step workflow for GPT-6 Astra: define goals, organize evidence, choose reasoning effort, verify results, and track costs.
¿Quieres probar los modelos de los que hablamos aquí? Registrarse