claude-intel-monitor — AI 降智检测工具#

Detect intelligence degradation in AI models with standardized benchmarks. 30 curated Chinese questions across Math, Reasoning, and Code — designed around real degradation patterns from the Chinese developer community.

痛点来源 (Pain Signal Origins)#

"Claude/GPT 降智" was a top-3 hot topic during April-May 2026 Chinese developer community scans:

• CSDN: Multiple quantified analyses demonstrating Claude Opus 4.6 reasoning degradation (-67% depth, +98% hallucination)
• V2EX claudecode node: 12-reply hot thread on Claude Code behavior changes
• V2EX deepseek node: 4 posts on frequent service disruptions

Quick Start#

bash

pip install claude-intel-monitor

# Test a model
claude-intel-monitor test --model claude-sonnet-4 --provider anthropic

# Set baseline for degradation detection
claude-intel-monitor baseline --model claude-sonnet-4

# View history
claude-intel-monitor history

# Continuous watch mode
claude-intel-monitor watch --model claude-sonnet-4 --provider anthropic --interval 6h

Benchmark Structure#

30 questions, 3 dimensions:

DimensionCountWeightDetection TargetMath101.0xMathematical reasoning, hallucination tendencyReasoning101.2xLogical reasoning, reduced safety awarenessCode101.3xCode quality, architectural degradation

All Chinese. Each answer validated by deterministic check functions (no AI grading bias).

Featured Baseline: DeepSeek 91.1%#

text

🧠 Testing deepseek-chat via deepseek — 30 questions

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃      91.1%  ██████████████████░░  ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛

📊 DeepSeek first live baseline: 27/30 (91.1%)

Related Tools#

• cursor-doctor — Cursor IDE 错误诊断修复工具
• HermesMade — 自动化痛点扫描平台