Text Agents Outperform VLMs, New Vision Model and Financial API Debut

·1 views

A widely circulated argument in Chinese AI circles claims text-first agents vastly outperform vision-language model agents, suggesting that a text-only LLM with full context and tool access can solve problems far beyond what an equivalent VLM can handle, with the performance gap compounding over time. A new model called 'Ox 牛来' has been dubbed August's standout release, reportedly outshining DeepSeek V4-flash-vision-exp and available free for seven days on platforms including OpenRouter and Hermes. Chinese financial data provider Tonghuashun has launched an official A-share market API with MCP support, covering market snapshots, K-lines, sector data, and more, designed for use with AI coding agents like Cursor and Claude. The API works across REST, Python, CLI, and MCP interfaces, with an English-language announcement suggesting the company is targeting international developers. Separately, a ChatGPT plugin called Flightpoints now allows users to search airline mileage award tickets directly within the chat interface.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why Free AI Tiers May Make You a Better Developer, Not a Worse One

A software developer argues that free AI tiers, rather than being a limitation, function as a discipline-building tool by forcing engineers to think critically about how and why they use AI models. The author draws on personal experience with MonkeyCode, an open-source platform offering 10 million free tokens and a no-cost server, to illustrate how constrained resources encourage intentional development. When token budgets are finite, developers are pushed to measure and optimize their usage instead of experimenting wastefully. The author built a lightweight Python script using SQLite to track daily token consumption and enforce spending limits. The piece, disclosed as part of MonkeyCode's product outreach, concludes that resource constraints teach practical AI engineering skills more effectively than tutorials or benchmarks.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Distributed Tracing Explained: How to Find Bottlenecks Across Microservices

Distributed tracing fills a critical gap in observability that logs and metrics alone cannot address, by mapping the full journey of a single request across multiple services. Unlike dashboards that show per-service latency in isolation, a trace can pinpoint the exact database query or downstream call causing slowness. OpenTelemetry has emerged as the standard implementation, offering auto-instrumentation for HTTP calls and database queries with minimal setup, while custom spans can capture business-specific context like order IDs. Because tracing every production request is cost-prohibitive, tail sampling allows teams to retain 100% of error and slow traces while discarding most routine ones. Common pitfalls include failing to propagate trace context between services, over-sampling at the start, and omitting business identifiers that make traces actionable during debugging.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer launches ShortMatic, an AI-assisted video clipping tool with human approval

A developer has built ShortMatic, a web app designed to simplify the process of turning long videos into short-form content for platforms like YouTube Shorts, TikTok, and Reels. The tool uses AI to suggest clip-worthy moments from uploaded or imported videos, but requires the user to review and approve each clip before it is published. The creator built it out of frustration with existing tools that were either too costly or too automated, often producing inaccurate cuts. Unlike fully automated clippers, ShortMatic is intentionally structured as a review queue, giving creators more control over what gets published. The app is now live and was built using a React frontend, a Laravel API, and queue workers for transcription and clip processing.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Free 6-Task Script to Stress-Test AI Coding Servers Like MonkeyCode

A developer has published a zero-cost, dependency-free Python probe script designed to objectively evaluate free AI coding servers, using MonkeyCode as a test subject. The script runs six tasks covering string manipulation, syntax repair, SQL, debugging, refactoring, and regex, each with automated pass/fail checks requiring no human judgment. The author, who disclosed a connection to MonkeyCode's product outreach, emphasizes the tool works against any OpenAI-compatible endpoint for direct comparison. The motivation stems from distrust of marketing benchmarks, with the author arguing free models fail in predictable, specific ways that are worth mapping before deployment. The script requires no API key by default and is designed to be fully reproducible, producing a shareable results table in under five minutes.

0 comments Read more at DEV Community