Model Releases

tldr: claude code changed some harness settings which degraded perf. these small harness tweaks can matter a lot! 1. default reasoning high …

tldr: claude code changed some harness settings which degraded perf. these small harness tweaks can matter a lot! 1. default reasoning high -> medium 2. bug that accidentally evicted thinking blocks o

DGX agentx-post
model-releasesharrison-chase--x

tldr: claude code changed some harness settings which degraded perf. these small harness tweaks can matter a lot! 1. default reasoning high -> medium 2. bug that accidentally evicted thinking blocks on every turn in session (march 26-april 10). was a change to help with cache optimization to evict old thinking blocks over an hour old 3. system prompt change to reduce verbosity which reduced code quality tbf it's easy to dogpile but the Claude Code team has a really hard job tracking and managing these small changes across the massive amounts of use-cases people use Claude Code for, not an easy spot to be in, the team is doing some superhuman work ❤️ but still this really shows the importance of: 1. Open Harnesses 2. Open Evals you want to be able to see and control the changes you're making to your harness and how they're justified across evals the power of open source + communities is that performance and security are better instead of worse because the entire world can inspect every change! will not be for everyone and that's fair but the idea of an open agent ecosystem is something we're excited to keep pushing on Thanks to the entire Claude community for giving feedback and continuing to build with us. Read the full post-mortem here: https://www.anthropic.com/engineering/april-23-postmortem

Related

Source: Harrison Chase (X) | 2026-04-23

Loading related sources…