Commenters debate whether increased reasoning effort in Muse Ember improves visual output quality, with some finding minimal settings superior and others noting persistent structural errors like incorrect leg positioning. The discussion also highlights skepticism about benchmark validity due to training data contamination causing convergent creative details across models.
140 readable replies · Supporting reply 1 · Supporting reply 2 · Supporting reply 3 · Supporting reply 4harbor-framework
harbor
Python2025–2025
harbor-framework / harbor
harbor
Framework for evaluating and improving agents
2Distinct people
2Mentions
0Explicit recommenders
7KGitHub stars
What to know before exploring
What it is
Framework for evaluating and improving agents
GitHub
Where it is used
No documented product use has been verified yet.
What people are saying
No recommendation phrase was found in the indexed comments.
No classified warning is attached to this page. Read the discussions before treating silence as approval.
Also in these discussions
No other repository is repeatedly linked in the same discussions.
This week on Hacker News
Nov 7–Nov 14 UTC · Direct repository links and separately labeled related project stories. Indexed through 11/14/2026, 1:49:46 PM UTC.
Mentions over time
HN mentions by 3-hour interval2 dated mentions
Observed mentions: 2025-11-08 10:49:49 UTC to 2025-11-11 09:59:36 UTC.
Counts come from the repository’s dated source evidence. Index cutoff: 2026-11-02 08:16:59 UTC.
View mention counts and date ranges
| Period | Start (UTC) | End (exclusive, UTC) | Mentions | Submissions | Comments |
|---|---|---|---|---|---|
| Nov 8, 09:00 | 2025-11-08 09:00:00 UTC | 2025-11-08 12:00:00 UTC | 1 | 1 | 0 |
| Nov 8, 12:00 | 2025-11-08 12:00:00 UTC | 2025-11-08 15:00:00 UTC | 0 | 0 | 0 |
| Nov 8, 15:00 | 2025-11-08 15:00:00 UTC | 2025-11-08 18:00:00 UTC | 0 | 0 | 0 |
| Nov 8, 18:00 | 2025-11-08 18:00:00 UTC | 2025-11-08 21:00:00 UTC | 0 | 0 | 0 |
| Nov 8, 21:00 | 2025-11-08 21:00:00 UTC | 2025-11-09 00:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 00:00 | 2025-11-09 00:00:00 UTC | 2025-11-09 03:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 03:00 | 2025-11-09 03:00:00 UTC | 2025-11-09 06:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 06:00 | 2025-11-09 06:00:00 UTC | 2025-11-09 09:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 09:00 | 2025-11-09 09:00:00 UTC | 2025-11-09 12:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 12:00 | 2025-11-09 12:00:00 UTC | 2025-11-09 15:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 15:00 | 2025-11-09 15:00:00 UTC | 2025-11-09 18:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 18:00 | 2025-11-09 18:00:00 UTC | 2025-11-09 21:00:00 UTC | 0 | 0 | 0 |
| Nov 9, 21:00 | 2025-11-09 21:00:00 UTC | 2025-11-10 00:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 00:00 | 2025-11-10 00:00:00 UTC | 2025-11-10 03:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 03:00 | 2025-11-10 03:00:00 UTC | 2025-11-10 06:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 06:00 | 2025-11-10 06:00:00 UTC | 2025-11-10 09:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 09:00 | 2025-11-10 09:00:00 UTC | 2025-11-10 12:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 12:00 | 2025-11-10 12:00:00 UTC | 2025-11-10 15:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 15:00 | 2025-11-10 15:00:00 UTC | 2025-11-10 18:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 18:00 | 2025-11-10 18:00:00 UTC | 2025-11-10 21:00:00 UTC | 0 | 0 | 0 |
| Nov 10, 21:00 | 2025-11-10 21:00:00 UTC | 2025-11-11 00:00:00 UTC | 0 | 0 | 0 |
| Nov 11, 00:00 | 2025-11-11 00:00:00 UTC | 2025-11-11 03:00:00 UTC | 0 | 0 | 0 |
| Nov 11, 03:00 | 2025-11-11 03:00:00 UTC | 2025-11-11 06:00:00 UTC | 0 | 0 | 0 |
| Nov 11, 06:00 | 2025-11-11 06:00:00 UTC | 2025-11-11 09:00:00 UTC | 0 | 0 | 0 |
| Nov 11, 09:00 | 2025-11-11 09:00:00 UTC | 2025-11-11 12:00:00 UTC | 1 | 0 | 1 |
What Hacker News says
2 sourcesFAfalcor84·Nov 11, 2025
I just saw that Terminal Bench introduced a new evaluation approach based on the new Harbor tool [0] and was surprised to see that it completely reshuffled the leaderboard, with the top 4 places now held by variants of gpt-5, whereas in terminal-bench@1.0 you had to scroll down to the 7th place to see gpt-5.
Does anyone here have any insight on whether this genuinely reflects capabilities better? I'm asking because last I checked, Codex+gpt-5 significantly underperformed Claude Code for my use case.
[0] https://github.com/laude-institute/harbor
In Terminal-Bench 2.0 and HarborHarbor – a framework for evaluating and optimizing agents and language models
3 points · Open discussion