Hackernews posts about Benchmarks

1. My personal AI benchmark: “Generate an SVG of a frog with a Habsburg jaw” (frogs.vaguespac.es)
   157 points by thebigship 18 days ago | 89 comments
2. GLM-5.3 Artificial Analysis Benchmarks (artificialanalysis.ai)
   150 points by apitman 2 days ago | 55 comments
3. When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation (arxiv.org)
   104 points by doppp 16 days ago | 131 comments
4. Goodhart's Law Comes for Every Benchmark You Trust (cacm.acm.org)
   100 points by pseudolus 20 days ago | 44 comments
5. Any text-to-SQL benchmark should address difficulties of real-world data stores (cacm.acm.org)
   62 points by shenli3514 29 days ago | 21 comments
6. Homebench – Benchmark local LLMs for speed, memory, and quality (github.com)
   59 points by davai-g 17 days ago | 8 comments
7. Enabling two settings tripled our scores on the ARC-AGI-3 benchmark (openai.com)
   38 points by tedsanders 22 days ago | 5 comments
8. Computer Anthology: A continuously evolving benchmark family for AI agents (vetto.ai)
   29 points by rigelbm 16 days ago | 11 comments
9. Spin audit of SQD/QSCI quantum-chemistry benchmarks on iron–sulfur clusters (zenodo.org)
   19 points by purestatelabs 14 days ago | 1 comments
10. Agentic test processes, LLM benchmarks, and other notes on agentic coding (danluu.com)
    17 points by bathtub365 26 days ago | 1 comments
11. Your model already knows the answer: how benchmark answers leak into LLMs (elman.ai)
    14 points by fran-mora 15 days ago | discuss
12. Show HN: Reproducibility Benchmark a Risk Quantitative Model (github.com)
    12 points by mingshi_tz 26 days ago | discuss
13. Show HN: Artificial Analysis tool to create custom benchmarks for any use case (artificialanalysis.ai)
    10 points by Gcam 7 days ago | discuss
14. Spin audit of SQD/QSCI quantum-chemistry benchmarks on iron–sulfur clusters (zenodo.org)
    9 points by purestatelabs 22 days ago | 6 comments
15. Modern-Fs-Benchmark (github.com)
    7 points by Grayskull 16 days ago | 1 comments
16. Show HN: Do Codex skills save tokens? A six-run task-size benchmark (codex-howto-benchmark.nguyenvantamdk2.chatgpt.site)
    7 points by phelannguyen164 18 days ago | 1 comments
17. Qwen 3.8 and Claude Opus 5 show why raw benchmark scores don't predict the bill (venturebeat.com)
    6 points by ashurandi 12 days ago | 1 comments
18. VAmoS Bench: Voice Agent Simulation Benchmark (veris.ai)
    6 points by _josh_meyer_ 20 days ago | 1 comments
19. Agents on Rails: the first benchmark report (rubyonrails.org)
    6 points by tosh 7 days ago | discuss
20. Show HN: Trunchbull, run real models against any benchmark in your browser (trunchbull.dev)
    6 points by pepperpoppins 8 days ago | discuss
21. Show HN: ExtractBench, an open-source schema extraction benchmark (github.com)
    6 points by cheesyFish 9 days ago | discuss
22. Goodhart's Law Comes for Every Benchmark You Trust (cacm.acm.org)
    6 points by alasr 23 days ago | discuss
23. Ask HN: Anyone interested in building a harness-only benchmark?
    5 points by GodelNumbering 16 days ago | 5 comments
24. Musk assembled a full-stack AI coding play while everyone watched benchmarks (pub.towardsai.net)
    5 points by glennall 6 days ago | 3 comments
25. Ouroboros: A coding agent that evolves its own harness and tops agent benchmarks (arxiv.org)
    5 points by ndreeew 10 days ago | 1 comments
26. A Lightweight Open-Source LLM Benchmark Tool – Compare Any Model on OpenRouter (cheikhhseck.medium.com)
    5 points by cheikhdev 16 days ago | discuss
27. PostgreSQL Autovacuum Internals and Benchmark (percona.community)
    5 points by nikolay_sivko 23 days ago | discuss
28. DeepSeek just dropped official V4 Pro GA benchmarks (twitter.com)
    4 points by binyu 8 days ago | 3 comments
29. Handwritten-edit benchmark: Fable 5 is #1, Opus 4.8 regresses 55% on miscounting (dorrit.pairsys.ai)
    4 points by svcrunch 28 days ago | 1 comments
30. 1Password's new benchmark teaches AI agents how not to get scammed (1password.github.io)
    4 points by mrkd 5 days ago | discuss