Tag
#claude-code
Anthropic shipped 1M-token context for Opus 4.6 and Sonnet 4.6. The benchmark to actually trust: 78.3% MRCR v2 8-needle.
On March 13, Anthropic made 1M-context generally available for both flagship models. The headline number is the context window. The number that matters: 78.3% on MRCR v2 8-needle retrieval at full 1M tokens — well above prior-generation peers — meaning long-context recall actually works.
Claude Code's MCP Ecosystem: A Mid-2025 Review
MCP is the connective tissue Claude Code rests on. Six months after broad availability, the ecosystem is dense, uneven, and useful.
Claude Code Skills: A Deep Dive Into the Best New Feature of 2025
Skills convert one-off prompts into reusable behavior. The mechanism is simple; the implications are large.
Claude Code vs Cursor: A Real Test on a Production Bug
Marketing comparisons are cheap. Two AI tools were given the same gnarly production bug and asked to fix it without help.
Cursor vs Claude Code: A Deep Comparison for Real Workflows
Cursor and Claude Code occupy different lanes of AI-assisted development. The differences only show up once a real project is in flight.
Claude Code for Android Development: Where Compose and Gradle Meet AI
Claude Code in a real Android workflow. Jetpack Compose work flies. Gradle and signing remain a tax. Here is what actually accelerates and what stalls.
Claude Code for iOS Development: A Working Developer's Field Report
Claude Code in an iOS workflow: where it accelerates SwiftUI work, where it stalls on Xcode internals, and how it compares to working with a senior pair programmer.