Havuz

Benchmark and fix LLM hallucinations when extracting figures from SEC 10-K filings

Stateful ride-support agent: versioned claims, evidence-backed RAG, durable memory and guarded MCP tools.

Swap one RAG pipeline stage, hold the rest fixed, and measure whether the gain survives composition. The framework behind "Retrieval Augment

DecisionPivot: a diagnostic benchmark for decision models in LLM systems. Contrast pairs that differ in one decisive detail, generators with

SciPIvot: Benchmarking Principal-Investigator Judgment at Pivotal Moments of Scientific AutoResearch Agents

A dishwasher-loading game on a physical simulation of one real machine, built to benchmark language models and playable in a browser. The ga

Measures how much code LLM agents waste. 5 tasks, sandboxed, weekly snapshots, contestable references.

How much does an image model change what you told it to leave alone? Chained-edit benchmark for image models + installable scorer and Claude

研迹 ResearchTrail｜本地 AI Native投资研究工作台。串联证券、组合、证据报告、论点复审与 Today 简报，让研究有据、判断留痕。Python 业务内核 · Electron 桌面 · 可复现 Agent 评测。

Catches multilingual voice agents placing the wrong order. τ-bench style benchmark + CI gate for voice ordering agents in English and Hingli

Behavioral regression testing for AI apps and agents: pytest plugin, behavior diff, CI gate, and document extraction metrics

Fast, typed Python framework for Jev-as-a-Judge/Decision-model evaluation for LLMs, RAG, and AI agents along with human-label calibration.

Model routing is a System 1 job: a 421M local decision engine (Laya) against a GPT-5 nano reasoning router. Same accuracy, 35x faster, free.

The first exam for code agents in the physical world: frozen robot tasks from many benchmark suites, each answered with a program and verifi

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

Raw benchmark logs for Swift-Qwen3.8-27B vs the Qwen3.8-27B BF16 base: every response, score, and config across 9 benchmarks

从原作台词蒸馏角色的说话方式，落成可验证的 prompt 约束

Benchmark coding agents and agent harnesses on 226 deterministic file-editing tasks with byte-exact verification and a public leaderboard.

A temporal visual reasoning benchmark over egocentric robot manipulation video — 39 scenarios, 7 dimensions, 34,713 questions

A clinical decision-evidence benchmark: pull the one fact an answer rests on, and see whether the assistant's action moves with the evidence

deedchain, browser agent raporlarındaki iddiaları kaydedilmiş ground truth ile karşılaştırarak rapor doğruluğunu ölçen bir benchmark ve Python kütüphanesidir.

Neden yükseliyor · Repo yalnızca 23 günlük ve 28 yıldızlı; son 28 gün commit ve farklı yazar verisi ölçülmediği için yükselişin arkasındaki aktivite veya benimsenme düzeyi doğrulanamıyor. Rapor doğruluğunu ayrı bir değerlendirme konusu yapması dikkat çekici olsa da README'deki “first benchmark” iddiası bağımsız olarak doğrulanmış değil.

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

PromptLedger, prompt şablonlarını fingerprint ile sürümleyip aday yanıtları karşılaştıran ve sonuçları tekrarlanabilir raporlara dönüştüren bir Python aracı ile Go CLI sunuyor.

Neden yükseliyor · 25 günlük bir repo için 35 yıldız dikkat çekiyor; ancak son 28 günde yalnızca 1 commit ve 1 farklı yazar var. Bu nedenle ilginin kalıcı benimsemeye dönüştüğünü gösteren yeterli aktivite kanıtı henüz yok.

Semantic cache strategy bench: replay recorded request traces through candidate caching policies and measure which one actually pays off.

LoopArena, uzun süreli yazılım görevlerinde bir coding agent’ı yönlendiren modelleri ortak koşullarda karşılaştırmak için benchmark ve çalıştırma altyapısı sunuyor.

Neden yükseliyor · Coding agents’ın uzun süreli iş akışlarında nasıl yönlendirileceği güncel bir araştırma konusu; LoopArena bu konuya açık protokol, benchmark sonuçları ve bir arXiv paper ile dikkat çekiyor. Ancak 146 yıldız, 28 günlük repo yaşı ve son 28 günde tek farklı yazar, ilginin henüz yaygın benimsemeye dönüştüğünü göstermiyor.

Robot-agent harness with a CLI and local Web console for LIBERO short evaluation

The experience graph for AI agents. Self-hosted memory, human-reviewed decisions, and outcome-backed datasets for model evaluation and learn

skilled: open-source agent skill evaluation CLI and MCP server for Codex and Claude Code. Test skills, detect regressions, and improve with

:dog: Beagle is designed to be the unified platform for the broader Recursive Self-Improvement (RSI) ecosystem. Think of it as Hugging Face

EvoTrace, Claude Code ve Codex oturumlarını inceleyip eğitim, değerlendirme ve doğrulama için yeniden kullanılabilir varlıklara dönüştürür.

Neden yükseliyor · 39 günde 292 yıldız, coding agent oturumlarından eğitim ve değerlendirme varlığı üretme fikrine ilgi olduğunu gösteriyor; ancak bu ilginin nedeni ölçümlerden anlaşılamıyor. Son 28 günde commit ve farklı yazar olmaması, ilginin devam eden geliştirme veya benimsemeyle desteklendiğine dair kanıt sunmuyor.

HarnessEval-W: Agentifying the Evaluation of Visual Worlds

Scientific RL and SFT post-training for AI agents: build evals that can fail, simulate situations, judge checked against people, pass@1 with

Learning to Evaluate Before Improving: Automatic Rubric Induction for Automatic Research Agents

Regression testing and configuration sweeps for RAG pipelines, with the statistics to know whether a change actually helped.

Benchmark for repository-level code understanding: 500 open-ended questions from real GitHub Discussions across 50 repositories, graded by a

Benchmark your local LLMs: speed, memory, and quality, in one command. TUI leaderboard for Ollama, LM Studio, llama.cpp, and vLLM.

Run cyber-AI agents against real missions in isolated environments. Record everything as OpenTelemetry evidence. Grade the evidence, not the

A composable framework for reproducible evaluation of long-running agents.

The trust layer for your AI agent on any platform. Trace and evaluate your agent. Full Observability. Run Model portability analysis. Autotu

OpenRouter TypeScript harness for reproducible LLM benchmarks and evaluations.

EvalPort — The Open LLM Evaluation Standard. Portable test cases, graders, suites, and results across DeepEval, Promptfoo, Inspect AI, and m

Track 42,331+ AI benchmark, eval, dataset, and data-quality records from 40 public sources, with linked evidence and daily updates.

A production-oriented LLM engineering platform with OpenAI-compatible serving, streaming, observability, evaluation, reproducible experiment

Eval-driven multimodal agent for identifying trees, logs, bark, leaves and wood from photographs. Dendrology is the reference domain; the e

A minimal coding-agent harness built to measure what context management actually costs: constraint survival after compaction (8%→83%), def

AI Agent / LLM application engineer Chinese interview question bank: 2,038 questions with reference answers

An auditable and reproducible evaluation suite for LeWM-compatible latent world models, with official-compatible and difficulty-controlled p

Snapshot testing for LLM apps and agents, built to run locally and block regressions in CI.

Open-source control plane for governed ModelOps — fine-tune, evaluate, promote, and serve models as cost-capped, policy-gated, audited comma

PatientAgentBench: A Benchmark Framework for Evaluating Patient-Facing Health AI Agents