Havuz

Physical-Engineering’s Last Exam

Cache-aware frequency sort: a header-only C++20 sort for low-cardinality integer arrays on x86-64.

Phone and word timestamp accuracy on human-labeled TIMIT and Buckeye, clean and under reverb, noise, music and babble. Two tracks, forced al

Laya (open source, local) vs TypeSafe Jev (API): two AI models race in Snake and fight in a Mortal-Kombat-style arena. Every move is a real

🛡️ Regex catches 0%, Meta's Prompt Guard 2 catches 1% of 629 realistic AgentDojo injection attacks when they're buried in tool output. Repro

Blind A/B testing for AI agent skills, .cursorrules, CLAUDE.md, and system prompts. Detects placebo and harmful instructions.

Compare web access APIs (web unblockers / scrape APIs) head-to-head against a fixed suite of ~90 real-world, hard-to-scrape URLs

A robot-oriented full-stack framework for long-horizon embodied agents in evolving physical and social environments.

Langauge2Drive: Do Vision-Language-Action Models Really Follow the Instructions in Autonomous Driving?

Piccolo: 4B, 9B and 27B decision models for agent tool review, context retention and data workflows, with an interactive Arena.

Benchmark and compare Ollama models across local and cloud endpoints with rich, sortable tables.

Verification-gated skill routing and self-improvement harness for Hermes-style agent skills

Multi-modal benchmark for measuring sensitive-information redaction in human recording data

Daily LLM value rankings - compare 300+ models by intelligence, speed and price. OpenRouter + Artificial Analysis. 大模型性价比排行榜

A public reference benchmark for radiology finding-to-report generation.

View planning with multi-turn VLM agents: ViewSuite 6-DoF benchmark on real ScanNet scenes + iterative RL-SFT training

Make every model cheaper or better. Zero-config behavioral harness for Claude Code, with a reproducible 408-run benchmark

Independent, evidence-based map of when TypeSafe's Jev actually holds up vs. breaks down — real API-call receipts, not a leaderboard. 中文為主的雙

Beyond Time Shifts (ECCV 2026): a reference-free audio-visual synchronization evaluator built on Qwen2.5-Omni-3B, with SynthSync data, R-GRP

TRACE Bench: Task-driven Roleplay Agentic Checklist Evaluation

The same HTTP API in 10 languages at 3 sizes, 30 implementations. Token cost for LLM agents, plus cold start, latency, throughput, CPU and m

A benchmark suite for differentiable physics solvers.

RigMark benchmarks local AI the way coding agents actually use it.

NarrowWide: A LiDAR-inertial dataset with repeated transitions between confined and open spaces (IEEE T-FR 2026)

Ten frontier VLM agents, five robotic capabilities, 1,000 cases. GeoProbe and traceable analyses of geometry, spatial reasoning, affordance,

SWE-bench for your codebase — mine your merged PRs into local, contamination-free coding-agent benchmarks. Adapters: claude-code, aider (Opu

WRBench: camera-controlled generation and diagnostic evaluation of video world models.

Give Vercel AI SDK agents a real persistent filesystem with Amazon S3 Files.

Run the AEON Bench suite on your own hardware: verified HuggingFace pull → serve → benchmark (text · agentic ×3 harnesses · vision · audio ·

Benchmarking coding agents on formalized undergraduate research problems

Logic-first AI code review via semi-formal execution tracing (Premises → Trace → Divergence → Trigger → Remedy). Catches behavioral bugs, ty

Community-driven benchmark suite for MLX inference engines on Apple Silicon

HoloGeo: Evidence-driven VLM for robust image geo-localization and landmark-bias mitigation.

AI-detector auditing toolkit: measures how often a detector flags genuine human writing, how stable its verdict is across seeds, and whether

⚽️🤖 Benchmarking LLMs and deep-research agents on real-world football prediction — from the tactical "who scores in minute 67" to the strate

A preservation-aware benchmark for natural-language SVG repair with deterministic binary rewards and complete model traces.

[ACM MM 2026] Official implementation of Self in Space: Benchmarking Self-Awareness and Spatial Cognition in UAV Embodied Intelligence

Real-time AI model power rankings (水墨战力榜) — usage, benchmarks, price-performance & speed from OpenRouter, as an ink-wash Streamlit dashboard

Open-source computational lithography workflow, manufacturability benchmarking, and foundation model data engine for advanced EUV/curvilinea

A reproducible benchmark for bimanual manipulation in simulation and the real world.

Reference harness for the Big Finance benchmark of workflow-grounded financial-research questions

CoderCup — the continuous public benchmark for AI coding agents, refereed by TestSprite

Advanced DNS benchmark (UDP, DoH, DoT) with scoring, latency tests and detailed reports

AgentBehaviorBench (ABB): A benchmark for evaluating AI agents’ ability to test other agents and detect behavioral defects.

An external memory for LLMs: a bi-temporal knowledge graph with hybrid (keyword + vector) retrieval and a self-improving reflection loop. Be

Interaction Bench: Benchmarked on what users feel

Benchmarks and notes for running modern LLMs with vLLM on 8x Tesla V100-32GB in 2026.

SWE-Serve: an agentic benchmark of 53 production inference-engineering tasks derived from merged SGLang pull requests, run with Harbor.

[ACM MM 2026 ORAL] PROVE: A Perceptual RemOVal cohErence Benchmark for Visual Media. Benchmark and removal metrics for image/video object r

PTCG-Bench: Can LLM Agents Master Pokémon Trading Card Game? (EMNLP 2026 Findings)