Top all-time · #17567
I
I've done some things
Personal blog · swelljoe.com/post · on HN as SwellJoeaiprogramming
I've done some things is a personal blog at swelljoe.com about ai and programming. Since 2026, 4 different Hacker News readers have linked it 26 times, most often “Will It Mythos?”. It is still writing: the last post was on 2026-07-24, and you can follow it by feed.
26mentions?
4HN people?
2026–2026first – last linked?
Too few sampled comments (3) to say how often HN recommends it.?
Why HN links it
Commenters cite these posts on open-model censorship and on testing agentic coding models, including a Qwen 3.5/3.6 MoE post-train and a 'will it mythos' evaluation of one model's bug-finding.
Summarised by Claude from a sample of 3 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
A personal blog
Most-linked posts?
- Will It Mythos?3 people · 10 mentions · first linked 2026-06-08
- /post/open-model-censorship/2 people · 4 mentions · first linked 2026-06-10
What Hacker News says?
Yeah, I pretty much had to switch to using GPT rather than Opus completely for all my security benchmarking and harness development. I was annoyed enough to blog about it: https://swelljoe.com/post/why-i-had-to-switch-to-gpt/
Comment linking /post/why-i-had-to-switch-to-gpt/
on “GPT-5.6”Submitted Will It Mythos? as a story · 31 points
HN discussionIn recent benchmarking I've been doing, DeepSeek V4 Pro was the fastest of 21 models, by a comfortable margin ( https://swelljoe.com/html/bench-report-final.html ). Faster than Claude Opus 4.8, which was the second fastest (Mistral doesn't count because it seems to have refused to participate). But, it's a limited data set, just a few benchmark runs of a limited set of tasks. It's entirely possible I happened to be…
Comment linking /html/bench-report-final.html
on “MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second”DeepSeek is the fastest model in the benchmarks I've been doing ( https://swelljoe.com/post/will-it-mythos/ ). Followed not so closely by Opus 4.8 and even less closely by Gemini 3.5 Flash and GPT 5.5. I've been really impressed with it, so far. It's also among the best at doing the work, though still trailing the frontier models from Anthropic and OpenAI.
Comment linking /post/will-it-mythos/
on “MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second”I've found in my current work on a security auditing harness and benchmarks, both Fable and Opus are useless. I recently switched to using GPT for Nelson and the security benchmarks I've been doing because Opus started refusing to do the work. I guess I probably could also use GLM or DeepSeek or MiMo, and I'll probably do some experiments to see the shape of all of their guardrails in this area soon, now that I see…
Comment linking /post/why-i-had-to-switch-to-gpt/
on “Fable is not a useful model”I did some more experiments and found DeepSeek V4 Flash actually answered pretty well on a bunch of other sensitive topics (Taiwan, the Great Firewall, Uyghurs, press freedom). It searched the web and fed back mostly the WikiPedia entries summarized. No complaints about those results, I'm pleasantly surprised. But, I asked Qwen 3.6 the same several questions and it is stridently pro-China. On Uyghurs: "The Chinese g…
Comment linking /post/open-model-censorship/
on “If Claude Fable stops helping you, you'll never know”I was surprised to find self-hosted DeepSeek V4 Flash answers accurately about almost every hot-button topic I could think of except Tiananmen Square, which it refused to answer. Self-hosted Qwen, on the other hand, is stridently supportive of the Chinese state. I posted the answers I got here https://swelljoe.com/post/open-model-censorship/
Comment linking /post/open-model-censorship/
on “Notes on DeepSeek”I have done replication tests of Qwen and the Gemma models. The Qwen benchmarks are published: https://swelljoe.com/post/qwen-quantization-degradation/ . (Though, I still want to add the other three cases to that one. I was mostly testing quantization effects in that test, but it also served as a replication test of Qwen in finding bugs.) The Gemma 4 replication tests are not published, yet, but Gemma 4 31B consiste…
Comment linking /html/gemma-promptlab-report.html
on “Will It Mythos?”I recently wrote up how I run local LLMs, because several folks had asked ( https://swelljoe.com/post/how-i-run-local-llms/ ) and I think even my setup, which I spent maybe $4200 on, half on a Strix Halo and half on upgrades for my desktop, would be too expensive to justify today. I bought before prices went through the roof, and only did so because I like to tinker with hardware...not because I expected it to ever…
Comment linking /post/how-i-run-local-llms/
on “Jamesob's guide to running SOTA LLMs locally”I have no idea. Seems like normal stuff. I used Kimi Code with K3 to add support for Kimi Code to flar ( https://swelljoe.com/post/i-let-every-agent-implement-its-ow... ), a task I've done with almost every major model/agent combo. Most show up as a blip on the usage chart...it's basically usually one file, a README update, and adding the agent name to the CLI. Then, I added it to my benchmark of security vulnerabil…
Comment linking /post/i-let-every-agent-implement-its-own-flar-backend/
on “Claude Opus 5”Maybe two things can be true at the same time? https://swelljoe.com/post/open-model-censorship/
Comment linking /post/open-model-censorship/
on “Apple to Skip High-End M6 Mac Chips in Favor of AI-Focused M7 Line”I added this to my benchmark of models looking for Mythos-reported security bugs. Unsurprisingly, it found 0. There is, after all, a lower bound on how small a model can be and still find security bugs. https://swelljoe.com/post/will-it-mythos/ It can seemingly reliably write working Python code though, which is impressive for such a little guy.
Comment linking /post/will-it-mythos/
on “VibeThinker-3B achieves 80.2 on LCBv6”I'm super enthusiastic about small models, but let's be realistic. A distillation is not the whole model (and, in fact, a lot of the small distillations on HuggingFace are worse than the base model...most of the Qwen 3.6 Opus/Fable/whatever distillations get weirder on some dimensions than Qwen 3.6 alone, as I understand it). There are little models that are very good for their size. I say nice things about Gemma 4…
Comment linking /post/gemma-4-exceeds-expectations/
on “I love LLMs, I hate hype”A 4-bit quantization of either Qwen 3.6 27b or Gemma 4 31b will run on a 32GB Mac with a decent-sized, but not full-sized, context. 64GB gets you the full ~256k context and you don't need to quantize your KV cache (though 8-bit quantization of KV may be worth it for performance). The 4-bit QAT version of Gemma 4 has practically identical performance to the full size version or the 8-bit version in most benchmarks an…
Comment linking /post/how-i-run-local-llms/
on “Kimi K2.7 Code is generally available in GitHub Copilot”Yes, it's "another post-train of Qwen 3.5/3.6 MoE intended for agentic use", according to https://swelljoe.com/post/will-it-mythos/ .
Comment linking /post/will-it-mythos/
on “Ornith-1.0: self-improving open-source models for agentic coding”Previously: https://news.ycombinator.com/item?id=48709744 https://swelljoe.com/post/will-it-mythos/ : "Poor performer here, only found the one bug that almost every model found, despite its performance on other benchmarks being excellent for its size. […] It also performs poorly in a chat without tools, exhibiting an ehthusiasm for hallucination. I’m currently working on a replication of this with full tool access,…
Comment linking /post/will-it-mythos/
on “Ornith-1.0: self-improving open-source models for agentic coding”The one that pleasantly surprised me when I did some testing was DeepSeek. The self-hosted version, at least, answered honestly about a wide variety of topics that are sensitive in China. The only topic it refused to discuss was Tiananmen Square. In every other case, it basically gave a WikiPedia summary. ( https://swelljoe.com/post/open-model-censorship/ ) And, I mean, censoring Tiananmen Square discussion is bad,…
Comment linking /post/open-model-censorship/
on “Where every major LLM stands politically”I added it to my benchmark based on Mythos-reported bugs, and it's better than GLM 5.1, but still behind several other models, maybe most directly comparable to Qwen 3.7 Max. But, several other open models, including small self-hostable ones (Gemma 4 and Qwen 3.6), found the same number of bugs, 3 of 9. Though it also gets partial credit for reporting one bug in the right spot, but kinda misunderstanding the bug. I…
Comment linking /post/will-it-mythos/
on “GLM-5.2 is the new leading open weights model on Artificial Analysis”Ah, the price has gone up on the AI Pro 9700, again, I see. So, yeah, it's a worse deal now. But, the V620 is not really a good deal, either, at $450 on eBay. It's roughly half the performance of the newer card. Usable but not blazing. And, you probably don't want to deal with cooling them; if you already have a 3D printer, you can print the shrouds for a few bucks and buy the fans for a few more bucks. But, if you…
Comment linking /post/how-i-run-local-llms/
on “We Have AI at Home Chapter 1: A Box of Scraps”This is really clever. Seems obvious in hindsight, as I've seen this tactic used for jailbreaks: modify the chat history to add the model affirming the user has the right to do the thing because they satisfied some requirement, and the model trusts itself to know the user is allowed to do the forbidden thing. But, also, Gemma 4 is really surprising on a bunch of fronts. It loses to Qwen 3.6 on most benchmarks, but i…
Comment linking /post/gemma-4-exceeds-expectations/
on “Do LLMs pass the mirror test?”I added this to a benchmark I've been doing of how well agents find security bugs, specifically security bugs originally found by Mythos. It performs poorly with only read/grep/ls tools, but in a follow-up test with a full shell and Python, it doubled its findings (still a poor showing, but it does at least indicate it is doing what it says on the tin: making tools to help it solve problems). It also did worse than…
Comment linking /post/will-it-mythos/
on “Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding”I tried adding GPT 5.5 Pro to a vulnerability scanning benchmark I made ( https://swelljoe.com/post/will-it-mythos/ ), and it blew through the $100 budget limit halfway through. DeepSeek V4 Pro cost about a dollar for the whole benchmark. GPT Pro cost an average of $22 per case (a case could be 1-5 files with a recent known vulnerability, usually just a single file and a prompt along the lines of "does this file hav…
Comment linking /post/will-it-mythos/
on “DeepSeek V4 Pro beats GPT-5.5 Pro on precision”I tried Kimi K3 on a task I've done with every other model I use regularly ( https://swelljoe.com/post/i-let-every-agent-implement-its-ow... ) and found it chewed a lot longer on the problem and ate up almost the entirety of a 5 hour usage limit on their $19 plan. I only have the $20 plan from OpenAI and the same task, with a lot of the same implementation details as Kimi Code, only took a few minutes and consumed a…
Comment linking /post/i-let-every-agent-implement-its-own-flar-backend/
on “The Kimi K3 Moment”I added GLM 5.2 to my security bug hunting benchmark when it came out, and found it to be a good performer, but not the best open model. The benchmark tests whether models can find bugs Mythos found. The best open models in the initial benchmark were DeepSeek V4 Pro or MiMo 2.5 Pro. But it turned out MiMo got lucky, it's performed worse on almost every test I've done since, while DeepSeek has consistently been among…
Comment linking /post/will-it-mythos/
on “Semgrep: GLM 5.2 beats Claude in our Cyber Benchmarks”It's terrible at hunting security bugs (I expected it to be, but I wanted to be sure). I added it to a benchmark I made with a corpus of some Mythos-discovered bugs, and it found zero. The smallest pretty successful models remain Qwen 3.6 and Gemma 4 (but I haven't tested the very small variants of those yet). https://swelljoe.com/post/will-it-mythos/
Comment linking /post/will-it-mythos/
on “VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO”