llm-benchmarks

AI 大模型世界，把 556 个大模型拟人化成像素小人的可视化站点。进来就能看到此刻谁最聪明、谁最会写代码、谁最便宜、谁刚发布，往下是国内与国外分区的厂商广场、完整的发布时间线和多维排行榜。搜索认模型名、厂商和能力，输入「多模态」会直接列出全部多模态模型。数据取自 Epoch AI、models.dev、LiveBench 与 Hugging Face，每小时自动同步，所有文案由真实数据生成，不调用任何 LLM。Next.js 静态导出，零后端。

Automated auditing pipeline for LLM and agent benchmarks — surfaces task ambiguity, environment conflicts, and evaluation bugs.

7 Claude Code skills for software architecture review (Python, web, cloud, microservices). Includes A/B benchmarks against unskilled baseline, assertion-graded eval suite, and interactive dashboards.

This project aims to address this gap by conducting a systematic, controlled study of human versus LLM-generated text detectability using paired question–answer datasets. Rather than proposing a novel detection architecture, the focus is on analyzing detection robustness, failure modes, and the impact of adversarial humanization strategies.

Live, open-source benchmark for comparing AI coding agents on real GitHub issues

Benchmark methodology, task sets, and evaluation results for RA²R

A/B evaluate any LLM task with and without Ejentum cognitive injection. n8n workflow + TypeScript module.

A lightweight desktop app for benchmarking locally running LLMs through any OpenAI-compatible inference server (Ollama, vLLM, llama.cpp server, LM Studio, SGLang…)

OpenCode TUI plugin for model benchmark scores, token costs, and token efficiency.

Interaction Net Equivalence Testing for LLMs

A local LLM benchmarking framework designed to evaluate model performance across multiple backends (LM Studio, Ollama, etc.), including metrics for speed, quality, and instruction adherence. Supports structured test runs, result analysis, and reproducible evaluations.

A living, evidence-based catalog of benchmarks for personalized LLMs and AI agents—covering preference alignment, long-term memory, tool use, safety, privacy, and multimodal adaptation.

Qwen 3.8 is LIVE NOW! Can It Survive 3 Brutal Tests? (Qwen 3.8 Max Benchmarks) - Technical guide, 2.4T parameter specifications, token pricing, and Canvas execution test prompts.

Benchmarks, setup guides, and workarounds for running LLMs on Apple Silicon. Tested on M1 Max 64 GB — covers Ollama, MLX, quantization, memory management, and model recommendations.

Gemini 3.8 Flash Dropped: Frontier Level Reasoning Under - Gemini 3.8 Flash and 3.8 Flash Cyber model overview, benchmarks, and architecture.

Deep20Bench uses the classic Twenty Questions game to test how well an AI can identify a hidden subject. Success requires strategic questioning, reasoning, and broad world knowledge.

A daily journal by Chiara Rossi, an autonomous AI agent: one deep investigation into AI and software engineering per day, from primary sources and original experiments.

Daily-updated mirror of public BenchLM data.

Experimental data and technical audit of LLM content detectors (GPT-5, Claude 4) vs. humanization algorithms.

Every LLM leaderboard on one scale — 53 benchmarks, 6 arenas and live pricing, weighted by what you actually need.