GLM-5.3-Flash and Qwen3.8-Flash Tested on 24 Real Tasks: Near-Identical Results

·1 views

A developer benchmarked two newly released open-weight AI models — GLM-5.3-Flash and Qwen3.8-Flash — against 24 real-world tasks drawn from an actual product stack, covering structured data extraction, SEO metadata generation, and code fixes. Both models were accessed via OpenRouter at near-similar costs, with GLM priced at $0.075 per million input tokens and Qwen at $0.15 per million. Across all three task categories, the models performed nearly identically in quality, with final scores falling within the noise floor of a 24-task sample. A notable practical difference emerged in token efficiency: GLM consumed roughly twice the output tokens as Qwen to produce equivalent results, suggesting higher reasoning overhead. The author concludes that for builders choosing between the two this week, prompt design and API reliability matter more than raw model intelligence.

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)

CS Student Builds AI Civic Issue Mapper to Track and Route Public Complaints

A computer science student developed a web platform called AI Civic Issue Mapper, allowing citizens to report civic problems such as potholes, garbage dumping, and broken streetlights using photos, GPS data, and descriptions. Built with Python Flask, MySQL, and JavaScript, the system automatically routes complaints to relevant departments, detects duplicates, and tracks resolution status. The project also includes a public transparency page where anyone can view complaint statistics without logging in, and was successfully deployed online. Throughout development, the student encountered real-world challenges including backend-database integration, authentication, email delivery, and deployment inconsistencies that tutorials had not prepared them for. The experience highlighted how building a functional application requires understanding how all system components interact, not just knowing individual technologies in isolation.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Unofficial open-source Grok bot ported to Linux by independent developer

An independent developer named jakob-bu has created an unofficial Linux port of the official Grok bot application. The project is open source and hosted on GitHub, making it freely accessible to the community. It was shared on Hacker News, where it received modest early attention with 9 points and one comment. The port aims to bring Grok bot functionality to Linux users who lack access to an official native version.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Engineer Uses Two AI Agents and 1 Billion Tokens to Design a Real Silicon Chip

A developer known as C Dragon completed two month-long projects using a two-AI, one-human team to design a SHA-256 cryptographic accelerator chip and integrate it into an open-source SoC. The workflow paired a planning AI (DeepSeek) with an executing AI (GLM) to separate design decisions from tool execution, reducing self-validation bias. The chip was built on SkyWater's 130nm open-source process and passed all 10 industrial signoff checks, including a transistor-level LVS match of 8,352 devices and zero DRC violations. Approximately one billion tokens were consumed across both projects, highlighting the significant computational cost of genuine AI-assisted hardware design. The author concludes that while AI proved capable of flow orchestration and error handling, the exercise also exposed clear boundaries to its reliability without structured human oversight.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Celmis launches self-hosted code intelligence tool with cross-repo symbol graph

Celmis is a newly introduced self-hosted code intelligence platform that indexes multiple repositories using a deterministic symbol graph built with tree-sitter. It allows developers to query codebases across language and repository boundaries, returning answers with precise file-and-line citations. The tool can review pull requests with awareness of callers in other repositories, audit dependencies, and generate tamper-evident SBOMs and findings archives with SHA-256 manifests. Celmis is also designed to serve its index over the Model Context Protocol, enabling editors and agents to access the same data under consistent access rules. Its release is partly motivated by the EU Cyber Resilience Act, which from September 2026 will require manufacturers to report actively exploited vulnerabilities within 24 hours, making early codebase visibility critical.

0 comments Read more at DEV Community