Claude Code's 33K-Token Load vs OpenCode's 7K: What the Gap Actually Means

·1 views

A logging proxy placed between coding agents and an AI model revealed that Claude Code sends roughly 33,000 tokens of scaffolding before any user prompt, compared to OpenCode's approximately 7,000 tokens. The bulk of Claude Code's token load comes from 27 tool definitions totaling nearly 100,000 characters, covering orchestration features like subagent delegation, worktree management, and monitoring — not just coding functions. OpenCode, by contrast, ships 10 tools and no scaffolding, making it leaner but also less capable out of the box. Testing showed the tradeoff is task-dependent: on complex multi-step work, Claude Code's batched tool calls made full sessions cheaper, while simple tasks farmed out to subagents ballooned token usage to over 500,000. The core question is not which baseline is smaller, but whether the extra capabilities justify their per-turn cost for a given workload.

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)

OTF Offers Shared Component API to Cut Cross-Platform UI Duplication

Cross-platform development teams routinely maintain separate UI components for web, iOS, and Android, leading to inconsistent designs and duplicated review cycles over time. OTF, a component library, addresses this by providing around 200 components under two package surfaces — one for web and one for iOS and Android — that share the same names and props. Rather than attempting a single runtime, OTF keeps the contract unified while letting each platform render using its own native system, such as SwiftUI on iOS and Compose on Android. The approach aims to eliminate drift between platforms that typically emerges months after launch, when one team adds a new variant and others fall behind. TypeScript enforcement is built in to catch any API mismatches across platforms during development.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Cuts Voice Agent Latency 5x by Streaming Sentences and Trimming Model Overhead

A developer building a local voice assistant found that fast LLM responses alone do not guarantee a smooth user experience, as delays accumulate across speech recognition, API calls, TTS synthesis, and audio playback. To address this, the developer modified the system so the backend streams text chunks to the device as the model generates them, rather than waiting for a complete response. The client splits incoming text into sentences using a regex boundary detector and immediately begins synthesizing each sentence, while a background worker overlaps audio playback with synthesis of the next sentence. Additional gains came from disabling extended model reasoning on conversational turns and only attaching the Google Search tool when the query actually required external information. The combined changes produced roughly a 5x reduction in time-to-first-byte on common chat turns and a noticeably more responsive perceived experience for the user.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Scrape Trip.com and Ctrip Hotel Reviews Using Python in 2026

Trip.com and its Chinese counterpart Ctrip (hotels.ctrip.com) host hotel reviews separately, meaning scrapers targeting only one platform miss a significant portion of guest feedback. Both sites use different page structures and languages, requiring developers to build two distinct scrapers and add a translation layer for Chinese-language reviews. Neither platform offers a public API, so extracting review data relies on reverse-engineering internal endpoints while navigating bot protection and rate limits. A DIY Python approach is feasible for single hotels but becomes complex at scale due to JavaScript-rendered pages and nested markup parsing. As an alternative, a third-party Apify actor called the Trip.com and Ctrip Reviews Scraper can process both feeds in one run, returning structured JSON with Chinese reviews pre-translated.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Git Turns 20: Why the Version Control Tool Remains Central to Software Development

Git has remained a foundational tool in software development for nearly two decades, quietly underpinning how code is written, reviewed, and shared across teams. The rise of GitHub extended Git's reach further, making open-source collaboration, pull requests, and CI/CD pipelines more accessible to developers worldwide. Even as AI accelerates code generation, Git's role has grown more critical as teams still need reliable ways to review, track, and understand code changes. However, many modern developer tools store project knowledge in proprietary platforms rather than directly within repositories, creating friction for collaborative workflows. Advocates argue that future developer tooling should be built around Git-native principles, keeping instructions and workflows versioned and accessible alongside the code itself.

0 comments Read more at DEV Community