GPT-4o (May)

Yigit Konur 公开的文档理解与处理案例，来源为 公开代码库，复核于 2026-06-26T21:50:00Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。

任务：这是一个围绕文档理解和结构化处理的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：Yigit Konur built api-llm-ocr, an open-source FastAPI service that uses GPT-4o's vision API to convert PDF documents into clean, structured markdown. The tool handles complex documents with tables, mixed layouts, header…

公开产物：公开材料提供公开代码、README 或项目配置，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：A production-ready tool that processes PDFs at ~1,500 tokens per page using GPT-4o vision. Cost: ~$15 per 1,000 pages with GPT-4o, ~$8 with GPT-4o mini, ~$4 with batch API. Supports parallel processing (50 pages in seco…

模型作用：GPT-4o (May) 在该案例中承担文档理解和结构化处理相关的生成、分析、编排或实现角色。 原始资料写作：GPT-4o's multimodal vision capability is the core differentiator — it reads and understands document context, not just character shapes like traditional OCR. The model correctly identifies table structures, headers, and…

A 类理由：有具体使用者、具体任务、公开原始证据和可访问产物，可公开核验。

风险边界：当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：Open-source GitHub repo (AGPL v3 license). GPT-4o explicitly listed in the cost comparison table. Demo video available in the repo. Author is an independent developer. Repo includes working code with configuration for O…

原始记录：api-llm-ocr: Production PDF-to-Markdown Conversion Using GPT-4o Vision

Be My Eyes 公开的多模态生成与理解案例，来源为 视频证据，复核于 2026-06-26。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。

任务：这是一个围绕多模态内容处理的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：Be My Eyes demonstrated GPT-4o assisting a blind user with real-time visual understanding in a live accessibility workflow.

公开产物：公开材料提供视频记录、页面说明或可访问产物，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：OpenAI launch page and Be My Eyes demo video provide the concrete model, user scenario and visible artifact.

模型作用：GPT-4o (May) 在该案例中承担多模态内容处理相关的生成、分析、编排或实现角色。 原始资料写作：OpenAI presented the Be My Eyes accessibility workflow as a GPT-4o multimodal capability demo.

A 类理由：有具体使用者、具体任务、公开原始证据和可访问产物，可公开核验。

风险边界：当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：This is a launch/demo artifact rather than a full production case study; verify current product routing before claiming ongoing GPT-4o use.

原始记录：Be My Eyes live accessibility demo with GPT-4o

Matthieu Le Cauchois 公开的多模态生成与理解案例，来源为 博客记录，复核于 2026-06-26T21:50:00Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。

任务：这是一个围绕多模态内容处理的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：Matthieu Le Cauchois built Shoggoth Mini, a soft tentacle robot that uses GPT-4o's real-time API for voice-driven interaction and high-level behavioral control. GPT-4o continuously listens to speech through the audio st…

公开产物：公开材料提供原始证据链接和可访问产物，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：A functional physical robot with expressive behaviors driven by GPT-4o's real-time audio API. The system has two control layers: low-level (hardcoded primitives + RL policies) and high-level (GPT-4o decision-making). GP…

模型作用：GPT-4o (May) 在该案例中承担多模态内容处理相关的生成、分析、编排或实现角色。 原始资料写作：GPT-4o's real-time API is the high-level brain of the system — it continuously processes the audio stream for speech recognition and makes contextual decisions about robot behavior. The model's ability to understand nat…

A 类理由：有具体使用者、具体任务、公开原始证据和可访问产物，可公开核验。

风险边界：当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：Blog post includes detailed technical documentation and video demonstrations of the robot in action. Author is a robotics engineer. The project is described as a hobby/research project, not a commercial product. GPT-4o …

原始记录：Shoggoth Mini: Expressive Soft Tentacle Robot with GPT-4o Real-Time Voice Control

Olup (product engineering team) 公开的多模态生成与理解案例，来源为 博客记录，复核于 2026-06-26T21:50:00Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。

任务：这是一个围绕多模态内容处理的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：Olup (a product engineer) used GPT-4o for production image detection to distinguish between 350 highly similar car illustrations in a museum setting. Users photograph an illustration on a wall, and the system must ident…

公开产物：公开材料提供原始证据链接和可访问产物，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：A deployed production system that accurately identifies which of 350 similar car illustrations a user has photographed. The pipeline combines a fast embedding-based KNN filter with GPT-4o for final disambiguation, balan…

模型作用：GPT-4o (May) 在该案例中承担多模态内容处理相关的生成、分析、编排或实现角色。 原始资料写作：GPT-4o serves as the final disambiguation step in a multi-stage vision pipeline. After KNN filtering narrows candidates from 350 to a handful, GPT-4o's vision capability performs fine-grained visual comparison to identi…

A 类理由：有具体使用者、具体任务、公开原始证据和可访问产物，可公开核验。

风险边界：当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：Blog post is on a Pages.dev hosted site (may be less permanent). HN discussion (222 points) confirms it is a real production deployment. The team is described as a 'crafty product engineer team' — likely a small agency …

原始记录：Production Image Detection for 350 Similar Museum Illustrations Using GPT-4o

Eduardo Blancas 公开的医疗与生命科学案例，来源为 博客记录，复核于 2026-06-26T21:50:00Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。

任务：这是一个围绕医疗和生命科学分析的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：Eduardo Blancas built an AI-assisted web scraper that uses GPT-4o's structured output feature to extract structured data from HTML tables and complex web pages. The system can parse complex table layouts (merged rows, n…

公开产物：公开材料提供原始证据链接和可访问产物，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：Successfully extracted structured data from complex HTML tables including: 10-day weather forecasts with nested day/night rows, Wikipedia tables with merged cells (Human Development Index), and hidden HTML elements that…

模型作用：GPT-4o (May) 在该案例中承担医疗和生命科学分析相关的生成、分析、编排或实现角色。 原始资料写作：GPT-4o's structured output feature (JSON mode) is the core of the system — it parses raw HTML into typed, schema-validated data structures. The model's vision-like understanding of HTML structure enables it to handle co…

A 类理由：有具体使用者、具体任务、公开原始证据和可访问产物，可公开核验。

风险边界：当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：Blog post includes demo links and source code references. Author is a data scientist (blancas.io). The project demonstrates both the power and cost limitations of GPT-4o for web scraping (377pts HN discussion). Blog pos…

原始记录：AI-Assisted Web Scraper: Structured Data Extraction and XPath Generation with GPT-4o

Tamer Comcuoglu 公开的研究与报告生成案例，来源为 博客记录，复核于 2026-06-26T21:50:00Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。

任务：这是一个围绕研究分析和报告生成的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：Tamer Comcuoglu used GPT-4o with LangChain's structured output (JSON mode) to classify and extract structured fields from over 10,000 'Ask HN: Who is Hiring' comments spanning May 2022 to June 2024. Each comment was par…

公开产物：公开材料提供原始证据链接和可访问产物，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：Successfully classified 10,891 job posting comments into structured records covering ~$54 in GPT-4o API costs. Produced actionable labor market insights: remote work trends, visa sponsorship rates, experience level dist…

模型作用：GPT-4o (May) 在该案例中承担研究分析和报告生成相关的生成、分析、编排或实现角色。 原始资料写作：GPT-4o was the core engine for structured data extraction, using JSON mode to output typed fields from unstructured HTML comment text. Its instruction-following capability enabled reliable field extraction including boo…

A 类理由：有具体使用者、具体任务、公开原始证据和可访问产物，可公开核验。

风险边界：当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：Blog post and HN discussion (465 points) are publicly accessible. Author is an individual developer, not a company. The GPT-4o model used is explicitly named. API costs ($54) are documented.

原始记录：HN Job Market Insights: Structured Analysis of 10,000+ Comments Using GPT-4o