Impressive paper showing how much the harness changes a coding agent's results.
This is a dev post classified by Jev as AI dev tools (a free resource), kept by the Dev Radar because it carries real work, not commentary.
Impressive paper showing how much the harness changes a coding agent's results. Harnesses do play a huge role in what you are getting out of the models. GPT-5.5 was run inside both Claude Code and Codex on the same 1,000 tasks. With a specialized PowerPoint workflow, it improved inside one harness and got worse inside the other. The harness also changed scores when the prompt was identical. ReFigBench asks coding agents to rebuild real arXiv overview figures as editable PowerPoint slides that keep the text, layout and connections. It covers ten configurations across the GPT, Claude, MiMo
Posted by elvis (320.8k followers) 43 min ago · 22 likes · 1.6k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: academy.dair.ai.
More dev work like this
- GPT-6 Sol & GPT-6 Luna by @OpenAI are now available on Design Arena! — @DesignArena
- Working in another AI app but need one of your Hermes agents? — @HermesWatcher
- Introducing Simple mode in Hermes Desktop. — @imbabybrooklyn
- Just built v prism on AI Studio with gemini 3.8, — @athrix_codes
- Action Gateway is the half of Managed Agents that handles tools. 16,000+ of them, from… — @digitalocean
- Luna is 100x cheaper than Astra. — @0x0SojalSec
- Omarchy is really good to test such useful things on Agents.👇 — @BkashJosi
- 处理长文本时，最麻烦的往往不是让 AI 总结，而是把里面的关键信息准确抽出来。 — @bkdgiffug
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 21.5k posts from 5k X accounts over the last 21 days, 2.4k tools, 12 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-22 20:31 UTC. Full method.