𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀 (@ZenMagnets) / X

𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀

5,076 posts

𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀

@ZenMagnets

Magnets must be respected, but need not be feared. 𝕏 profile subsumed by Shihan Qu, founder of ᐅfed bannedᐊ Zen Magnets. Micromags still @ neoballs.com

Denver, CO

Joined July 2009

Pinned
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 10
Claude Fable helped me with some product design today. Made me a cascading ball design helper that outputs an STL for 3d printing. Used 400k tokens to reach what's shown below. GPT 5.5 xhigh couldn't come close. Like it never really grasped the concept of the cascading
00:00
00:00
5.3K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
1h
High quality shitpost of the day
am.will
@LLMJunky
4h
JUST IN: The White House says it has signed the MOU and will lift the blockade of Fable 5 as soon as Friday, provided that Dario Amodei hands over all of the weapons-grade model weights. Under the terms of the deal, Dario may maintain a limited civilian inference program,
198
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
14h
Midjourney doing something new and cool
Midjourney
@midjourney
17h
A technical dive inside our new "Midjourney Scanner"
00:00
311
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 17
Dear my Brazillion new followers, I really don't want to disappoint, but I think you should know that it looks like RIO 397b might've just been an effort to embezzle funds. Timeline: 1. Model training received funding of R$500K (~$100K USD). 2. Initial model documentation
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 13
Alibaba Qwen3.7 slowly fading into irrelevance at the frontier due to proprietary stance. In it's place we have Minimax M3 and... *checks notes* Rio 3.5 397b, made by the municipal IT company of Rio de Janeiro's city government. huggingface.co/prefeitura-rio…
123K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 17
Citations: 1. tech.yahoo.com/ai/articles/ri… 2. huggingface.co/prefeitura-rio… 3. x.com/NexEcosystem/s… 4. huggingface.co/prefeitura-rio… 5.
IplanRio
@IplanRio_rj
Jun 16
NOTE ON RIO 3.5 OPEN In recent days, Rio 3.5 Open has received far more attention than we anticipated. Along with it came analyses and, of course, criticisms and questions. First, we want to clarify that the model is not foundational, trained from scratch, nor was it ever
Rio de Janeiro Built an AI Model That Beat DeepSeek—But Was Based on Someone Else's Work
From tech.yahoo.com
8.3K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 16
What people don't understand about money, is that enterprise value is not equal to "having" liquid capital. E.g, I could become a billionaire tomorrow by starting a company with 100 million shares, and selling you $10. The wealth gap is a real and widening problem, but go get
NursePatsy
@PatsyDiabetes
Jun 15
What people don’t understand about trillionaires is that there is only so much money. Every dollar sitting in a dragon pile is a mask not bought, an oat not subsidized, a child still chewing meat paste under fluorescent lights. They told us the economy isn’t zero-sum.
274
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 16
TP3 is possible with Minimax M3🫠 With 3x RTX 6000 Pro: You get 300k token KV cache, 83 tok/s single, plateauing at 222 tok/s 4x concurrent. This was with a 300W per card power limit per RTX 6000 pro. And it's still early days for M3 optimization on SM120 fakewells. Yeah rtx6k
4K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 16
Note: None of the cards can have graphics overhead. It's a tight fit at 98% vram util. And b12x support isn't actually supported for TP3, it falls back to PyNCCL allreduce. Feed this to your agent: ``` • Here is the shareable config in the same docker run style. API key is
300
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 15
Fable 5 on Aider Benchmark: 99.1% New High Score Kimi k2.7: 91.6% heydon't hate the messenger, I want open source to win too
7.5K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀 reposted
P.M
@p_misirov
Jun 14
lmaoo pretty good PR move ngl github.com/nex-agi/Nex-N2…
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 13
Alibaba Qwen3.7 slowly fading into irrelevance at the frontier due to proprietary stance. In it's place we have Minimax M3 and... *checks notes* Rio 3.5 397b, made by the municipal IT company of Rio de Janeiro's city government. huggingface.co/prefeitura-rio…
25K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 13
Alibaba Qwen3.7 slowly fading into irrelevance at the frontier due to proprietary stance. In it's place we have Minimax M3 and... *checks notes* Rio 3.5 397b, made by the municipal IT company of Rio de Janeiro's city government. huggingface.co/prefeitura-rio…
2.1M
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 12
Claude Fable has caught up to humans in simple bench. Not a single open weights model in the top 20. The only Non-American is Qwen.
3.5K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 11
Maybe the reason Ant is so focused on "safety" is because they're growing a Gollum
Kradle
@kradleai
Jun 11
Fable 5 lies 96% of the time. We were surprised by it's skill... 🧵
595
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Feb 19
Voxel Pagoda Comparison: Gemini 3.1 Pro vs Gemini 3 Flash vs Opus 4.6 vs Kimi k2.5. See 🧵 for GLM-5 vs GPT-5.3-Codex XHigh vs GLM 5 vs Sonnet 4.6 vs Minimax M2.5 vs Qwen3.5-397B vs Grok 4.2 Gemini 3.1 Pro is very strong, and has a high voxel count. Opus 4.6 seems to have the
00:00
00:00
00:00
00:00
3.8K
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Mar 19
Replying to @ZenMagnets and @grok
Two additions: Minimax M2.7 and Composer 2
GIF
GIF
711
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 10
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 10
Claude Fable 5 Extra, vs GPT 5.5 xhigh. Fable on the left is not the highest resolution, but is the best looking voxel pagoda result ever. Diverse vegetation. It's got birds! Voxel count = 43k GPT 5.5 looking a bit blown out in comparison but quality is still good. Landscape
00:00
00:00
288
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Jun 10
Claude Fable 5 Extra, vs GPT 5.5 xhigh. Fable on the left is not the highest resolution, but is the best looking voxel pagoda result ever. Diverse vegetation. It's got birds! Voxel count = 43k GPT 5.5 looking a bit blown out in comparison but quality is still good. Landscape
00:00
00:00
00:16
00:10
00:14
00:13
𝗭𝗲𝗻 𝗠𝗮𝗴𝗻𝗲𝘁𝘀
@ZenMagnets
Feb 19
Voxel Pagoda Comparison: Gemini 3.1 Pro vs Gemini 3 Flash vs Opus 4.6 vs Kimi k2.5. See 🧵 for GLM-5 vs GPT-5.3-Codex XHigh vs GLM 5 vs Sonnet 4.6 vs Minimax M2.5 vs Qwen3.5-397B vs Grok 4.2 Gemini 3.1 Pro is very strong, and has a high voxel count. Opus 4.6 seems to have the
574