API-QuickCheck

API-QUICKCHECK 2.0

█████╗ ██████╗ ██╗ ██████╗ ██╗ ██╗██╗ ██████╗██╗ ██╗ ██╔══██╗██╔══██╗██║ ██╔═══██╗██║ ██║██║██╔════╝██║ ██╔╝ ███████║██████╔╝██║ ██║ ██║██║ ██║██║██║ █████╔╝ ██╔══██║██╔═══╝ ██║ ██║▄▄ ██║██║ ██║██║██║ ██╔═██╗ ██║ ██║██║ ██║ ╚██████╔╝╚██████╔╝██║╚██████╗██║ ██╗ ╚═╝ ╚═╝╚═╝ ╚═╝ ╚══▀▀═╝ ╚═════╝ ╚═╝ ╚═════╝╚═╝ ╚═╝ ───────────────────────────────────────────────────────────── ⚡ Universal AI Relay Forensics, Batch Testing & Agent Adapter

🚀 Live Demo · Why QuickCheck · The 9 Flagship Models · 24-Probe Suite · Batch Testing · Quick Start · 🇨🇳 中文说明

⚡ Why API-QuickCheck 2.0?

In today's AI API relay and proxy marketplace, model substitution fraud (e.g. routing requests to cheap quantized checkpoints, fake <think> reasoning tags, silent context truncation, and proxy-level rate-limiting) is rampant.

API-QuickCheck 2.0 is an industrial-grade forensics engine designed to verify the true identity, streaming quality, and agent capability of your model endpoints against the 9 latest frontier models:

• 🛡️ Cryptographic Proof: Anthropic official private-key Thinking Signature turn-2 verification;
• 🧠 Thought Stream Forensics: Native reasoning_content delta validation exposing hardcoded fake reasoning tags;
• ⚡ Microsecond Precision: Measures TTFT (Time to First Token), live TPS throughput, and Chunk Jitter variance;
• 🔑 Industrial Batch Testing: ±25% Jitter anti-ban scheduling, domain endpoint memory caching, and balance sniffing;
• 🔄 Autonomous 2026 Baselines: 3-day automated CI/CD synchronization with authoritative model catalogs;
• 🔒 Zero Data Logging: Crafted in the Anthropic Claude Warm Dark Editorial Design System, with 100% in-memory client execution.

🏆 The 9 Frontier Models Supported

API-QuickCheck 2.0 is calibrated against the 9 most powerful models shaping modern agent workflows across OpenAI, Anthropic, Google, and xAI:

┌─────────────────────────┬───────────────────────────────┬─────────────────────────────────────────────────────────────┐
│ Provider                │ Model ID                      │ Frontier Tier & Core Forensics Focus                        │
├─────────────────────────┼───────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ 🟢 OpenAI               │ gpt-5.6-sol                   │ Flagship Frontier · Hard Tool Planning, Code Repair, 64K    │
│                         │ gpt-5.6-terra                 │ Balanced Workhorse · Responses API & Strict JSON Schema     │
│                         │ gpt-5.6-luna                  │ Sub-Second High-Throughput · Low Latency Agent Execution   │
├─────────────────────────┼───────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ 🟠 Anthropic            │ claude-fable-5                │ Next-Gen Frontier · Adaptive Thinking Signature Continuity │
│                         │ claude-opus-5                 │ Frontier Coding & Deep Reasoning · Complex Tool Chains      │
│                         │ claude-sonnet-5               │ Balanced Agent Workhorse · Messages API & Prompt Caching    │
├─────────────────────────┼───────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ 🔵 Google               │ gemini-3.1-pro-preview        │ 2M Multimodal Frontier · Thought Signature & PDF/Chart Ext  │
│                         │ gemini-3.7-flash              │ Ultra-Fast Agent Orchestrator · Interactions API Routine    │
├─────────────────────────┼───────────────────────────────┼─────────────────────────────────────────────────────────────┤
│ ⚫ xAI                  │ grok-4.6                      │ Realtime Frontier · Controlled Tools, Reasoning & Sandbox   │
└─────────────────────────┴───────────────────────────────┴─────────────────────────────────────────────────────────────┘

🎯 24-Probe Forensics & Capability Matrix

Every model is audited against our 24-Probe Balanced Suite across 4 critical layers:

                                 ┌── [ P0 Protocol (7) ] ➔ Model Discovery, Native Route, Auth, SSE Stream, Strict JSON, Tool Shape, Invalid Param
                                 ├── [ P1 Architecture (5) ] ➔ Reasoning Config, State Continuity, Tool Roundtrip, Signature Continuity, Cache Semantics
API-QuickCheck 2.0 ──────────────┼── [ P2 Capability (8) ] ➔ Constraint JSON, Tool Planning, Code Repair (A/B), Vision Chart, Needle Context (Start/Mid/End)
                                 └── [ P3 Quality (4) ] ➔ Cold Start TTFT, Concurrency TPS, Network Jitter, P95 Stability Samples

1. Deep Fidelity & Downgrade Forensics

• Anthropic Official Private-Key Thinking Signature Check (p1-signature-continuity):
  Captures the cryptographic private-key signature inside the thinking block, then injects it back in turn 2. If the relay routed to a downgraded or counterfeit model, the second turn will fail cryptographic verification instantly!
• Native Thought Stream vs. Injected Text (p0-native-route):
  Differentiates between true SSE reasoning_content delta events and text <think> tags spoofed by middleman proxies.
• Metacognitive & 64K Needle In A Haystack Probes (p2-context-start/middle/end):
  Tests deep context retrieval at 64,000 tokens across start, middle, and end positions to catch silent context truncations.

2. High-Precision Streaming Speed Benchmark

• Microsecond TTFT (Time to First Token): Accurately isolates network queueing and handshake latency from model compute time.
• Real-Time Token Throughput (TPS): Live velocity tracker with visual terminal typewriter animation.
• Jitter Variance Matrix: Measures inter-chunk delivery variance to reveal proxy rate-limiting and buffer stalling.

🔑 Batch API-Key Testing & Anti-Ban Pool

Test hundreds of API Keys concurrently with enterprise-grade resilience:

• Smart Dirty Text Extraction: Cleans messy pastes from plain text, commas, CSV columns, and JSON arrays.
• Adaptive Anti-Ban Protection:
  • ±25% Jitter Delay: Desynchronizes concurrent requests to bypass middleman rate-limit thresholds;
  • Domain Endpoint Memory Pool: Caches working endpoint routes to reduce probing traffic by 80%;
  • 429 Circuit Breaker: Automatically backs off upon gateway congestion.
• Universal Quota Sniffer: Multi-endpoint balance parser supporting OneAPI, NewAPI, DoneAPI, OpenRouter, SiliconFlow, DeepSeek.
• Multi-Format Export: 1-click export to TXT, CSV (Excel UTF-8 BOM compliant), JSON with 20-batch history rollback.

📊 Technical Comparison Matrix

🚀 Quick Start

Option 1: Live Web App (No Installation Needed)

Visit the live deployment directly: https://api-quick-check.vercel.app/

Option 2: Local Development

# 1. Clone repository
git clone https://github.com/som1ng/API-QuickCheck.git
cd API-QuickCheck

# 2. Install dependencies
npm install

# 3. Start local development server
npm run dev

# 4. Run 24-probe audit test suite
npm test

# 5. Sync latest 2026 model baselines
npm run sync:models

🛠️ Tech Stack

• Core Engine: React 18, TypeScript, Vite
• Design System: Anthropic Claude Warm Dark Editorial System, TailwindCSS
• Rendering & Math: ReactMarkdown, KaTeX, Rehype-Katex, Mermaid.js
• Icons: Lucide React
• Edge Deployment: Vercel Serverless Edge Functions

📄 License

Distributed under the MIT License. See LICENSE for more information.