Hacker News: guregodevohttps://news.ycombinator.com/user?id=guregodevoHacker News RSShttps://hnrss.org/hnrss v2.1.1Sun, 01 Nov 2026 12:05:40 +0000<![CDATA[New comment by guregodevo in "Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard"]]>For this benchmark i 've been using OpenRouter. If you have enough RAM or you can ssh a cloud shell this intro https://memdoor.ai/docs/open-weight can work too. I have reached with the ssh door a Google Cloud Box where the 27B model ran with an L4 g2-standard-16. A inference engine server is setup easily with memdoor (https://github.com/guregodevo/janis) on any CPU or GPU box. Ideally i think we need a quorum of SSH boxes like this, distributed that you can ping with a mention such as @box:coder. It's WIP . Any suggestion for running a team of workers in remote machine is welcome. I'd like to find a cheap solution.

]]>Sun, 01 Nov 2026 11:50:33 +0000https://news.ycombinator.com/item?id=50324747guregodevohttps://news.ycombinator.com/item?id=50324747https://news.ycombinator.com/item?id=50324747<![CDATA[New comment by guregodevo in "Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard"]]>This is a benchmark that solves ten hard problems in five passes. It scores 9.8 against the authors' 9.6 and Fable 5's 8.0. Memdoor workflows are the harness that replicates the original GVS5H thread harness posted in HN last week. Raw results (i.e. each call's tokens, cost and host) are in the bench folder for verification.

]]>Sun, 01 Nov 2026 09:11:58 +0000https://news.ycombinator.com/item?id=50323833guregodevohttps://news.ycombinator.com/item?id=50323833https://news.ycombinator.com/item?id=50323833<![CDATA[New comment by guregodevo in "Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard"]]>Happy hallowen! ::::

Actually we tried Cerebras. It caps output at 32K tokens so the 128K reasoning (the author budget) didn't fit. Cerebras speed is real though

]]>Sun, 01 Nov 2026 00:35:55 +0000https://news.ycombinator.com/item?id=50321660guregodevohttps://news.ycombinator.com/item?id=50321660https://news.ycombinator.com/item?id=50321660<![CDATA[New comment by guregodevo in "Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard"]]>will definitively try it. Overall with a smarter model the improvement is in the token usage and latency. Tbh I think i could get a 10 score if i tuned the prompt and temperature a little more. The original author harness is already giving 9.6 score which is pretty good. However this harness loop can take more than 1 hour on the hardest problem at 100 token /sec. A smarter model such as 3.8-flash-next could get 10 score and resolve the hardest problem faster. That's my estimate i'll give it a try. Thx

]]>Sun, 01 Nov 2026 00:22:32 +0000https://news.ycombinator.com/item?id=50321592guregodevohttps://news.ycombinator.com/item?id=50321592https://news.ycombinator.com/item?id=50321592<![CDATA[New comment by guregodevo in "Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard"]]>Here the intent is to replicate GVS5H with memdoor a headless coding agent I created months ago. Anyone can reshape the harness by redefining the steps in yaml files , models and temperatures. Up to you guys to try you have the recipe in this repo.

]]>Sun, 01 Nov 2026 00:05:07 +0000https://news.ycombinator.com/item?id=50321492guregodevohttps://news.ycombinator.com/item?id=50321492https://news.ycombinator.com/item?id=50321492<![CDATA[Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard]]>Article URL: https://github.com/guregodevo/memdoor/tree/main/examples/workflows/team-solve

Comments URL: https://news.ycombinator.com/item?id=50320995

Points: 271

# Comments: 123

]]>Sat, 31 Oct 2026 22:46:12 +0000https://github.com/guregodevo/memdoor/tree/main/examples/workflows/team-solveguregodevohttps://news.ycombinator.com/item?id=50320995https://news.ycombinator.com/item?id=50320995<![CDATA[New comment by guregodevo in "Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness"]]>I tried to compare with a workflow i built with this harness : https://github.com/guregodevo/memdoor/tree/main/examples/wor...

The main issue i found is that Qwen3.8-27B thinks a lot. On a hard problem the planner runs 32K tokens of reasoning and never writes the plan. Even with thinking off on Cerebras, a problem takes roughly 3 minutes and the 100 problems about 1.5 hours.

]]>Sun, 25 Oct 2026 20:48:20 +0000https://news.ycombinator.com/item?id=50235925guregodevohttps://news.ycombinator.com/item?id=50235925https://news.ycombinator.com/item?id=50235925<![CDATA[New comment by guregodevo in "Show HN: Memdoor – A coding agent that can build and run headless workflows"]]>A worflow is basically a DAG of tasks that are defined in YAML files. Once the agent has built it it can be run retried and repeated. Think of workflows as a DAG in Airflow or any DAG orchestrator.

]]>Wed, 21 Oct 2026 15:33:42 +0000https://news.ycombinator.com/item?id=50182753guregodevohttps://news.ycombinator.com/item?id=50182753https://news.ycombinator.com/item?id=50182753<![CDATA[Show HN: Memdoor – A coding agent that can build and run headless workflows]]>I wrote a coding agent that is on par with most coding agents and is particularly well designed for building and running repeatable steps that automate your long running tasks. I experienced many times that skills are not fitted for daily task such data job backfills, CI steps and complex DAGs etc..

Comments URL: https://news.ycombinator.com/item?id=50180337

Points: 1

# Comments: 1

]]>Wed, 21 Oct 2026 13:10:14 +0000https://github.com/guregodevo/memdoorguregodevohttps://news.ycombinator.com/item?id=50180337https://news.ycombinator.com/item?id=50180337<![CDATA[New comment by guregodevo in "Show HN: CodingBrain – acquire skills faster"]]>Pretty intuitive tool. I like it.

]]>Wed, 28 Oct 2015 15:51:49 +0000https://news.ycombinator.com/item?id=10465100guregodevohttps://news.ycombinator.com/item?id=10465100https://news.ycombinator.com/item?id=10465100