Marés de modelos

Notas sobre levar IA em streaming para produtos reais: clientes, modelos, custos e protocolos.

2026-10-25
- Marés de modelos · ZH
  WebSockets 与 SSE：逐步搭建 AI 聊天流式响应
  从消息方向、浏览器接入到断线恢复，逐步比较 WebSockets 与 SSE，搭建支持逐词元输出、取消生成和可靠重连的 AI 聊天应用。
- Marés de modelos · EN
  Gemini streamGenerateContent with alt=sse: Step by Step
  Build a Gemini SSE stream step by step, from an authenticated request to event parsing, incremental rendering, and production checks.
- Marés de modelos · EN
  Pay-as-You-Go vs Monthly Passes for Chat Products
  A step-by-step guide to comparing token costs, monthly passes, usage limits, and margins as your streaming AI chat product grows.
2026-10-24
- Marés de modelos · ZH
  开放权重 Qwen 模型接入流式技术栈：从部署到逐字输出
  逐步搭建开放权重 Qwen 模型的流式调用链，覆盖推理服务、兼容接口、服务端转发、前端渲染、取消请求与上线监控。
- Marés de modelos · ZH
  Claude Sonnet 5.5 工具调用实战：逐步搭建智能体工作流
  以订单查询为例，逐步讲解 Claude Sonnet 5.5 的工具定义、流式调用、结果回传与安全控制，并通过 Fluxio 接入实时智能体应用。
- Marés de modelos · ZH
  Gemini 4 Argon 生产聊天接入：分步实践与注意事项
  从接口验证、流式渲染到会话管理、故障恢复和灰度发布，逐步梳理 Gemini 4 Argon 接入生产聊天的实施方法与注意事项。
- Marés de modelos · ZH
  实时翻译功能中小模型与大模型的选择
  分步骤比较小模型与大模型，设计兼顾速度、成本和译文质量的实时翻译功能。
- Marés de modelos · EN
  Fable vs Opus: Choose a Premium Chat Flagship
  A step-by-step guide to comparing Fable and Opus on answer quality, streaming latency, token costs, and premium chat reliability.
- Marés de modelos · EN
  Which Model Streams Code Best in Chat?
  Follow a practical walkthrough to compare coding models for streaming speed, code quality, cost, and a smoother chat interface.
- Marés de modelos · EN
  Context Windows and Streamed Replies: A Practical Guide
  Learn how context windows affect streamed AI replies, budget tokens, handle output limits, and build reliable real-time experiences.
- Marés de modelos · EN
  How to Pick Models for a Voice Assistant Pipeline
  Choose voice assistant models step by step, balancing latency, streaming, tool accuracy, speech quality, and cost with realistic tests.
- Marés de modelos · EN
  MiniMax and Hunyuan Fallbacks for Streaming Chat
  Build reliable streaming chat with MiniMax M3 and Hunyuan 4 fallbacks, safe retry rules, clear timeouts, and production monitoring.
- Marés de modelos · EN
  Grok for Conversational Products: A Practical Guide
  Build a streaming Grok chat experience step by step, from API setup and conversation state to safety, latency testing, and launch.
- Marés de modelos · EN
  GLM and Qwen for Multilingual Live Chat
  Build multilingual live chat with GLM and Qwen, streaming responses, reliable language handling, and practical production safeguards.
- Marés de modelos · EN
  Build Kimi Long-Context Chats With Streaming
  Build a streaming Kimi chat interface with secure API calls, context budgeting, incremental rendering, and reliable error handling.
- Marés de modelos · EN
  Build Budget DeepSeek Chatbots That Stream Smoothly
  A step-by-step guide to building a budget DeepSeek chatbot with token streaming, secure API access, cancellation, and cost controls.
- Marés de modelos · EN
  Gemini 3.1 Pro vs Claude Opus 5.5: Reasoning Guide
  Compare Gemini 3.1 Pro and Claude Opus 5.5 with a step-by-step test for reasoning quality, multi-turn reliability, streaming latency, and cost.
- Marés de modelos · EN
  Gemini Flash Live Summaries: A Step-by-Step Guide
  Build live summaries with Gemini Flash, streaming responses, rolling context, and a practical update loop through an OpenAI-compatible API.
- Marés de modelos · EN
  Opus vs Sonnet for Streaming Chat: Step by Step
  Compare Opus and Sonnet for streaming chat with practical steps for latency testing, quality evaluation, routing, and Fluxio integration.
- Marés de modelos · EN
  GPT-6.1 Sol for Low-Cost Streaming Assistants
  A step-by-step guide to building a fast, token-efficient streaming assistant with GPT-6.1 Sol and Fluxio.
- Marés de modelos · EN
  GPT-6 Astra vs GPT-5.6 Sol: Long-Form Workflow
  Compare GPT-6 Astra and GPT-5.6 Sol for long-form writing with a practical workflow covering prompts, streaming, quality checks, and cost.
- Marés de modelos · EN
  When Haiku Is Enough for Real-Time Autocomplete
  A step-by-step guide to choosing Claude Haiku 4.5 for fast, useful autocomplete without unnecessary model cost.