Parsimony – an open-source benchmark for how much code AI models add
An open-source benchmark of how much code AI models add.
Parsimony – an open-source benchmark for how much code AI models add launched on October 13, 2026 with 77 votes, #181 of 2,243 launches that month and more than 95% of that year's launches.
In plain words
Parsimony is an open-source benchmark that measures code efficiency in AI models by counting syntax elements added and removed, rather than lines of code. It evaluates how much excess code AI generates when solving programming tasks, ranking models by mean net units added, with deletions subtracted. The tool addresses a gap in existing benchmarks that focus on solve rate and cost but ignore code bloat, which impacts long-term codebase maintainability.
written from the facts on this page · October 2026
From the sources
Existing benchmarks measure solve rate and cost, but don't punish excess code, which makes a codebase harder to work with in the long term. Parsimony measures how much code AI models add. It takes the patches that benchmarks already publish and counts coding units : syntax elements like statements, calls, comparisons, names and literals. Lines aren't counted, so cramming code isn't rewarded, and deleted code is subtracted: net units = added − removed. Models are ranked by mean net units added on the tasks they solved, lowest first. "Net units added (mean)" also includes failed attempts. Solved comes from SWE-bench Verified ; tests aren't rerun. Measured is how many solved attempts could be…from parsimony.lasu.dev
Does a similar job
all alternatives →
Kilo Code for VS CodeMay 2026 · kilo.ai · ▲896Parallel agents, diff reviewer, and multi-model comparisons
Weave2025 · app.workweave.ai · ▲237Hey HN, We’re building Weave: an ML-powered tool to measure engineering output, that actually understands engineering output! Why? Here’s the thing: almost every eng leader already measures output - either openly or behind closed doors. But they rely on metrics like lines of code (correlation with effort: ~0.3), number of PRs, or story points (slightly better at ~0.35). These metrics are, frankly, terrible proxies for productivity. We’ve developed a custom model that analyzes code and its impact directly, with a far better 0.94 correlation. The result? A standardized engineering output…
ChunkHound, a local-first tool for understanding large codebasesJan 2026 · github.com · ▲114ChunkHound’s goal is simple: local-first codebase intelligence that helps you pull deep, core-dev-level insights on demand, generate always-up-to-date docs, and scale from small repos to enterprise monorepos — while staying free + open source and provider-agnostic (VoyageAI / OpenAI / Qwen3, Anthropic / OpenAI / Gemini / Grok, and more). I’d love your feedback — and if you have, thank you for being part of the journey!
Xcessiv – Fully managed web application for automated machine learning2017 · github.com · ▲108A web-based application for quick, scalable, and automated hyperparameter tuning and stacked ensembling in Python. - reiinakano/xcessiv
More ai this month
the category →
Launched alongside, October 2026
the whole month →- IV
Life & fun · 7d ago · tanzaniaeducationproject.org
Turn a phone, tablet or TV into a sign. Type a message and it fills the screen: welcome signs, timers, Wi-Fi passwords and more. No app, no account, free.
Life & fun · 8d ago · bigwords.page
Play a free virtual guitar online with your keyboard or by touch. Slow songs down and see every chord, string and strum. Acoustic, classic, electric and bass.
Life & fun · 2d ago · guitoro.com
Carrier settings from iPhone, Pixel and Galaxy firmware, decoded and compared: APNs, VoLTE, 5G and Wi-Fi Calling per carrier, and what each build changed.
Life & fun · 6d ago · carrierexplode.com