71/100Not Verifiable
YouTube·News·
Cursor ditches VS Code, but not everyone is happy...
by Fireship
View original on YouTube →
Summary
Cursor 3.0, a complete rewrite in Rust, moves beyond a VS Code fork to an AI agent management platform, aiming for a "zero code future." It introduces Composer 2, an in-house coding model initially claimed to surpass Claude Opus but later revealed to be based on Moonshot's Kimi K2. The new interface allows users to run swarms of AI agents in parallel across various environments, significantly accelerating development by automating code generation and design fixes.
IntermediateCoding AssistantsAgentsModel ReleaseOpen Source
Benchmark Cross-Reference
How the creator's claims compare to independent benchmark data
CursorNo Data
Creator says: Major upgrade but misleading claims about model performance(mixed)
Claude OpusSupported
Creator says: Used as benchmark comparison(neutral)
Rank #4Intelligence: 53$10.00/Mtok41 t/s
Benchmarks show: Claude Opus 4.6 ranks #4 on AA (intelligence 53) and #4 on BenchLM (score 92), supporting use as benchmark comparison
Moonshot Kimi K2Supported
Creator says: Revealed as actual base model behind Composer 2(neutral)
Rank #14Intelligence: 47$1.20/Mtok35 t/s
Benchmarks show: Kimi K2.5 shows intelligence index 47, ranking #14 - confirms it's a real model that could be base for Composer 2
Sources: artificialanalysis.ai, aider.chat, benchlm.ai
Tools Discussed
Cursor
Major upgrade but misleading claims about model performance
Claude Opus
Used as benchmark comparison
Moonshot Kimi K2
Revealed as actual base model behind Composer 2
Score Breakdown
Raw score: 71= 71/100
AI Quality Analysis
31 / 40Originality6
Specificity6
Completeness5
Value Density7
Honesty Limitations7
Model: anthropic/claude-sonnet-4
Context Signals
13 / 20Freshness4
Author Track Record2
Genuine Engagement7