Loading video...

Video Failed to Load

Go Home

Parameter scaling just broke in front of me. I gave 90 math problems as images to 10 local vision models, 3 runs each, only consistent answers across all 3 runs counted. Two findings: > Gemma 4 was the most consistent family, 31B took the crown at 89.6%. > But...

stevibe's profile picture

stevibe

22,057 subscribers

22,882 views • 2 months ago •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

3 ways to destroy a piece of paper. Qwen 3.5 35B A3B vs. Ornith 1.0 35B, side-by-side canvas test. (Why 3.5 not 3.6? Ornith is post-trained on Qwen 3.5 and Gemma 4, so this shows what the post-training adds.) Same 3 challenges: 🔪 Slice: three blade swipes, fruit-game style 📄 Shredder: desktop strip-cut 🗑️ Crumple: balled up and tossed Winner: not close. Ornith, decisively. The post-training quality is REAL.
0:20

3 ways to destroy a piece of paper. Qwen 3.5 35B A3B vs. Ornith 1.0 35B, side-by-side canvas test. (Why 3.5 not 3.6? Ornith is post-trained on Qwen 3.5 and Gemma 4, so this shows what the post-training adds.) Same 3 challenges: 🔪 Slice: three blade swipes, fruit-game style 📄 Shredder: desktop strip-cut 🗑️ Crumple: balled up and tossed Winner: not close. Ornith, decisively. The post-training quality is REAL.

stevibe

47,792 views • 1 month ago

So we know Gemma 4 is good at tool calling, but what about web coding? I threw 4 UI screenshots at three Gemma 4 models and said rebuild this, one shot, no hand-holding, just image in, code out. Model lineup: - E4B - 26B A4B (MoE) - 31B Dense (skipped the E2B this round) Let me know which one you think cooked the hardest
0:21

So we know Gemma 4 is good at tool calling, but what about web coding? I threw 4 UI screenshots at three Gemma 4 models and said rebuild this, one shot, no hand-holding, just image in, code out. Model lineup: - E4B - 26B A4B (MoE) - 31B Dense (skipped the E2B this round) Let me know which one you think cooked the hardest

stevibe

124,942 views • 4 months ago

Most people think a 0.8B model is a toy. 3rd test. This time: OCR I fed Qwen 3.5 0.8B three images: a Kindle page, my terrible handwriting x 2. Mostly correct. Running locally. From a model smaller than your browser tab.
0:33

Most people think a 0.8B model is a toy. 3rd test. This time: OCR I fed Qwen 3.5 0.8B three images: a Kindle page, my terrible handwriting x 2. Mostly correct. Running locally. From a model smaller than your browser tab.

stevibe

125,136 views • 5 months ago

a new 8GB VRAM GPU dense Local LLM leader was born yesterday runs on: RTX 4060 / RTX 3070 / RTX 2080. any 8GB card Qwen 3.5 9B (dense) was the go to for 6-8GB VRAM builds. Gemma 4 12B QAT (dense) just changed that. same llama.cpp + cuda 13.2. i7 12700H. 16GB RAM. same -ngl 99 flags. same 48k context. unsloth gemma-4-12b-it-Q4_K_M.gguf → 15 tok/sec @ 48k ctx unsloth gemma-4-12B-it-qat-UD-Q4_K_XL.gguf → 32 tok/sec @ 48k ctx → 26 tok/sec @ 64k ctx 64k context is a big deal. Hermes 3 agent requires 64k minimum to run. you're now getting full hermes compatible context on a budget consumer GPU at 26 tok/sec locally. 2.1x faster on identical hardware. and here's the part that breaks your brain: the QAT-UD-Q4_K_XL is actually SMALLER than the Q4_K_M "XL" why? QAT = Quantization Aware Training Google didn't train the model first and compress it later they trained it to be quantized from day one the weights already know how to survive low precision that's why you get more quality per byte llamacpp flags: -m gemma-4-12B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 48000 -v fits in 8GB VRAM clean. no API. no cloud. no subscription. and this isn't even the MTP variant yet Gemma-4-E2B QAT runs on 3GB RAM, E4B on 5GB, 12B on 7GB, 26-A4B on 15GB and 31B on 18GB. I have benchmarked the 26b and 31b qat as well on a single RTX 4090, checkout the comments for details. If you have a 6GB or 8GB VRAM GPU, post your numbers. more benchmarks and configs coming soon
0:22

a new 8GB VRAM GPU dense Local LLM leader was born yesterday runs on: RTX 4060 / RTX 3070 / RTX 2080. any 8GB card Qwen 3.5 9B (dense) was the go to for 6-8GB VRAM builds. Gemma 4 12B QAT (dense) just changed that. same llama.cpp + cuda 13.2. i7 12700H. 16GB RAM. same -ngl 99 flags. same 48k context. unsloth gemma-4-12b-it-Q4_K_M.gguf → 15 tok/sec @ 48k ctx unsloth gemma-4-12B-it-qat-UD-Q4_K_XL.gguf → 32 tok/sec @ 48k ctx → 26 tok/sec @ 64k ctx 64k context is a big deal. Hermes 3 agent requires 64k minimum to run. you're now getting full hermes compatible context on a budget consumer GPU at 26 tok/sec locally. 2.1x faster on identical hardware. and here's the part that breaks your brain: the QAT-UD-Q4_K_XL is actually SMALLER than the Q4_K_M "XL" why? QAT = Quantization Aware Training Google didn't train the model first and compress it later they trained it to be quantized from day one the weights already know how to survive low precision that's why you get more quality per byte llamacpp flags: -m gemma-4-12B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 48000 -v fits in 8GB VRAM clean. no API. no cloud. no subscription. and this isn't even the MTP variant yet Gemma-4-E2B QAT runs on 3GB RAM, E4B on 5GB, 12B on 7GB, 26-A4B on 15GB and 31B on 18GB. I have benchmarked the 26b and 31b qat as well on a single RTX 4090, checkout the comments for details. If you have a 6GB or 8GB VRAM GPU, post your numbers. more benchmarks and configs coming soon

Alok

259,993 views • 2 months ago

nemotron 3 omni q8 on dgx spark 128gb vram cranking via hermes agent at 56 tok/s. first night of real local agentic on this box and local hits harder than i thought it would. q8 (near lossless quant, perplexity loss <1% vs fp16) running 256k context on 33 gb of unified memory, 90+ gb still free. multimodal omni weights included. hermes agent driving from telegram, talking to it from bed. speed: 56 tok/s generation, 1,300 tok/s prefill. for context, qwen 3.6 27b at q4 (heavy quant) on 3090 = 40 tok/s. nemotron at higher precision quant on spark beats qwen at lower precision quant on 3090. moe 3.5b active params architecture earns its keep. what i tested tonight: agentic tool calling works clean. ask it to check disks, it autonomously runs df -h through hermes agent. ask it to set up telegram gateway, it invokes the hermes-agent skill, walks through the prompts, completes the flow. overthinks a bit before tool calls (reasoning model trait) but lands the right move every time. researches api docs, internalizes, tests, documents. completes tasks. current models on dgx spark: 9 gguf files, 305 gb total, mix of qwen 3.6 27b dense (5 quants), nemotron omni (4 quants), deepseek v4-flash 158b q4 (the 112gb flagship test). more data coming this week as i benchmark each.
1:40

nemotron 3 omni q8 on dgx spark 128gb vram cranking via hermes agent at 56 tok/s. first night of real local agentic on this box and local hits harder than i thought it would. q8 (near lossless quant, perplexity loss <1% vs fp16) running 256k context on 33 gb of unified memory, 90+ gb still free. multimodal omni weights included. hermes agent driving from telegram, talking to it from bed. speed: 56 tok/s generation, 1,300 tok/s prefill. for context, qwen 3.6 27b at q4 (heavy quant) on 3090 = 40 tok/s. nemotron at higher precision quant on spark beats qwen at lower precision quant on 3090. moe 3.5b active params architecture earns its keep. what i tested tonight: agentic tool calling works clean. ask it to check disks, it autonomously runs df -h through hermes agent. ask it to set up telegram gateway, it invokes the hermes-agent skill, walks through the prompts, completes the flow. overthinks a bit before tool calls (reasoning model trait) but lands the right move every time. researches api docs, internalizes, tests, documents. completes tasks. current models on dgx spark: 9 gguf files, 305 gb total, mix of qwen 3.6 27b dense (5 quants), nemotron omni (4 quants), deepseek v4-flash 158b q4 (the 112gb flagship test). more data coming this week as i benchmark each.

Sudo su

30,228 views • 3 months ago

I gave two MoE models the same vibe coding challenge Qwen3.6 35B A3B (31.8GB) vs Gemma4 26B A4B (23.3GB) Stack: &gt; Unsloth Q6_K_XL &gt; llama.cpp &gt; Model-card recommended sampling for each 4 prompts, side-by-side. Which one do you think wins?
0:38

I gave two MoE models the same vibe coding challenge Qwen3.6 35B A3B (31.8GB) vs Gemma4 26B A4B (23.3GB) Stack: &gt; Unsloth Q6_K_XL &gt; llama.cpp &gt; Model-card recommended sampling for each 4 prompts, side-by-side. Which one do you think wins?

stevibe

262,265 views • 3 months ago

Round 2: Dense edition Qwen3.5 27B vs Gemma4 31B Same 4 HTML canvas challenges, this time no MoE. Setup: > Quant: Unsloth Q6_K_XL (both) > Each running on an RTX 5090 > Single HTML file, no libraries, real-time animation My read per prompt: 🌳 Tree — Gemma, branches are beautifully structured ✨ Night sky — close, but Qwen's stars blink better 🐟 Fish boids — Qwen, detailed fish and look at that motion, so realistic 🏠 House build — neither really nailed it Agree? Does dense actually beat MoE here?
0:43

Round 2: Dense edition Qwen3.5 27B vs Gemma4 31B Same 4 HTML canvas challenges, this time no MoE. Setup: > Quant: Unsloth Q6_K_XL (both) > Each running on an RTX 5090 > Single HTML file, no libraries, real-time animation My read per prompt: 🌳 Tree — Gemma, branches are beautifully structured ✨ Night sky — close, but Qwen's stars blink better 🐟 Fish boids — Qwen, detailed fish and look at that motion, so realistic 🏠 House build — neither really nailed it Agree? Does dense actually beat MoE here?

stevibe

33,304 views • 3 months ago

I just crammed the updated Gemma 4 26B A4B QAT (MoE) with 180k context into an 8GB RTX 4060 (8 GB VRAM + 16 GB RAM only!!) and optimized the batch size. 23 tokens/sec decode, 300 tokens/sec prefill Yesterday I showed you a Gemma 4 31B dense model running flawlessly on an RTX 4090. Today, we're breaking the VRAM bank on a budget card using Unsloth’s new Gemma 4 26B (A4B) QAT quants. Following Google’s chat template update that boosted agentic benchmarks by +10%, I pushed this model to its absolute limits. Here is how you squeeze 250k context out of 8GB of VRAM. # The Setup & The Optimization - Hardware: Nvidia RTX 4060 (8GB VRAM) + 16GB System RAM - Environment: CUDA 13.0 build of llama.cpp - Model: gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf - Prompt: 28,000 tokens of prompt for each run If you read my L2 cache breakdown (attached in replies), you know the 4060’s 24MB cache maxes out at `-b 1024 -ub 1024`. Push past that, and prefill crashes. I locked those flags in for every test below to ensure maximum GEMM throughput. # 1. The Raw Context Push (Unquantized KV Cache) First, I wanted to see how far pure 8GB VRAM + 16GB RAM could stretch without touching the KV cache: - 80k Context: Prefill 385 t/s | Decode 25.5 t/s - 120k Context: Prefill 270 t/s | Decode 24 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 120000 --port 8080 -ub 1024 -b 1024 Without KV quantization, 120k is your hard ceiling. push past that prefill throughput drops off a cliff, making the model practically unusable for large agentic workloads. # 2. The Q8 KV Cache Lifeline To survive 250k context on a budget card, you have to quantize the KV cache. I enabled 8 bit KV cache (`-ctk q8_0 -ctv q8_0`) and re ran: - 180k Context: Prefill 280 t/s | Decode 22.8 t/s - 250k Context: Prefill 115 t/s | Decode 20 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 180000 --port 8080 -b 1024 -ub 1024 -ctk q8_0 -ctv q8_0 Result: Q8 KV cache brings 250k context back from the dead. Decode speed stabilizes at a highly usable 20 t/s. You are trading a very small bit amount of reasoning precision for an extra 130,000 tokens of context window. if you own a single rtx 3050, 3060, 3070, 4050, 4060, 5050 or 5060, you must try this model and optimize your batch size for higher prefill. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. What model are you running on your 6GB, 8GB or 12GB cards right now? Let's see your setups.
0:20

I just crammed the updated Gemma 4 26B A4B QAT (MoE) with 180k context into an 8GB RTX 4060 (8 GB VRAM + 16 GB RAM only!!) and optimized the batch size. 23 tokens/sec decode, 300 tokens/sec prefill Yesterday I showed you a Gemma 4 31B dense model running flawlessly on an RTX 4090. Today, we're breaking the VRAM bank on a budget card using Unsloth’s new Gemma 4 26B (A4B) QAT quants. Following Google’s chat template update that boosted agentic benchmarks by +10%, I pushed this model to its absolute limits. Here is how you squeeze 250k context out of 8GB of VRAM. # The Setup & The Optimization - Hardware: Nvidia RTX 4060 (8GB VRAM) + 16GB System RAM - Environment: CUDA 13.0 build of llama.cpp - Model: gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf - Prompt: 28,000 tokens of prompt for each run If you read my L2 cache breakdown (attached in replies), you know the 4060’s 24MB cache maxes out at `-b 1024 -ub 1024`. Push past that, and prefill crashes. I locked those flags in for every test below to ensure maximum GEMM throughput. # 1. The Raw Context Push (Unquantized KV Cache) First, I wanted to see how far pure 8GB VRAM + 16GB RAM could stretch without touching the KV cache: - 80k Context: Prefill 385 t/s | Decode 25.5 t/s - 120k Context: Prefill 270 t/s | Decode 24 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 120000 --port 8080 -ub 1024 -b 1024 Without KV quantization, 120k is your hard ceiling. push past that prefill throughput drops off a cliff, making the model practically unusable for large agentic workloads. # 2. The Q8 KV Cache Lifeline To survive 250k context on a budget card, you have to quantize the KV cache. I enabled 8 bit KV cache (`-ctk q8_0 -ctv q8_0`) and re ran: - 180k Context: Prefill 280 t/s | Decode 22.8 t/s - 250k Context: Prefill 115 t/s | Decode 20 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 180000 --port 8080 -b 1024 -ub 1024 -ctk q8_0 -ctv q8_0 Result: Q8 KV cache brings 250k context back from the dead. Decode speed stabilizes at a highly usable 20 t/s. You are trading a very small bit amount of reasoning precision for an extra 130,000 tokens of context window. if you own a single rtx 3050, 3060, 3070, 4050, 4060, 5050 or 5060, you must try this model and optimize your batch size for higher prefill. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. What model are you running on your 6GB, 8GB or 12GB cards right now? Let's see your setups.

Alok

36,617 views • 20 days ago

Qwen 3.8 Max vs Opus 5, side-by-side. 🌀 The Inception top, spin, wobble, fall 💡 A hand-crank generator lighting a bulb Qwen 3.8 Max vs Claude Opus 5. Both rendered. One costs 8x more. Worth it? Watch and decide.
0:22

Qwen 3.8 Max vs Opus 5, side-by-side. 🌀 The Inception top, spin, wobble, fall 💡 A hand-crank generator lighting a bulb Qwen 3.8 Max vs Claude Opus 5. Both rendered. One costs 8x more. Worth it? Watch and decide.

stevibe

49,166 views • 9 days ago

First impressions on Muse Glimmer! It's incredibly fast for a dense model, currently running an average of 208tps with a max of 274tps on a single 5090 with their DFLASH config. Comparatively, though, both using Open Code, Qwopus Coder (with thinking off) produced a much better shark survival game than the one I got from Glimmer. Meta's new dense model is currently just lacking some HTML canvas taste, but this is something that can be added via SFT as long as the model is stable and capable from a back-end programming perspective. And it seems to be, without a doubt. The big kicker here is that I ran this at extra high thinking, and it did not take long at all to run. Our current local leader, Qwen 27B 3.6, has a tendency to overthink, but with glimmer, that is not the case. Right now, my recommendation for general local programming (Apps, Games, Websites, Visual Tools) in this class is still Qwopus Coder with thinking disabled, or Qwopus Fusion with thinking enabled. Of course Shark Survival is a very basic domain-specific test, but I find that the result scales very well across many domains. If we're going to be shipping apps generated entirely locally, visual taste is somewhat of a bare minimum requirement, solely in my opinion, and Qwen's models in this class offer significantly more at the moment. That's actually why I initially started getting into finetuning with Qwen 3.5, they were the first base that was able to do really good front-end with some opus-trace fine-tuning. Qwen 3.6 has taste even in the base model, and we know Qwen 3.8 is going to blow us all away! Regardless, this looks like a very tempting new base model. As a first offering from Meta in this class for a long time, I am incredibly impressed and elated to have it. We now finally have a proper Single GPU frontier race, instead of us just begging Qwen for more releases. Single GPU open frontier model race is a VERY good thing. Please keep pushing Meta
1:20

First impressions on Muse Glimmer! It's incredibly fast for a dense model, currently running an average of 208tps with a max of 274tps on a single 5090 with their DFLASH config. Comparatively, though, both using Open Code, Qwopus Coder (with thinking off) produced a much better shark survival game than the one I got from Glimmer. Meta's new dense model is currently just lacking some HTML canvas taste, but this is something that can be added via SFT as long as the model is stable and capable from a back-end programming perspective. And it seems to be, without a doubt. The big kicker here is that I ran this at extra high thinking, and it did not take long at all to run. Our current local leader, Qwen 27B 3.6, has a tendency to overthink, but with glimmer, that is not the case. Right now, my recommendation for general local programming (Apps, Games, Websites, Visual Tools) in this class is still Qwopus Coder with thinking disabled, or Qwopus Fusion with thinking enabled. Of course Shark Survival is a very basic domain-specific test, but I find that the result scales very well across many domains. If we're going to be shipping apps generated entirely locally, visual taste is somewhat of a bare minimum requirement, solely in my opinion, and Qwen's models in this class offer significantly more at the moment. That's actually why I initially started getting into finetuning with Qwen 3.5, they were the first base that was able to do really good front-end with some opus-trace fine-tuning. Qwen 3.6 has taste even in the base model, and we know Qwen 3.8 is going to blow us all away! Regardless, this looks like a very tempting new base model. As a first offering from Meta in this class for a long time, I am incredibly impressed and elated to have it. We now finally have a proper Single GPU frontier race, instead of us just begging Qwen for more releases. Single GPU open frontier model race is a VERY good thing. Please keep pushing Meta

Kyle Hessling

16,191 views • 2 days ago

I designed a new test specifically for multimodal models: fill out a paper form. And it's much harder than it sounds. This isn't typing into an electronic field that captures your text. The form is just an image. The model has to place each form element: text, checkmarks — at the correct pixel position on the canvas itself. Results: 🟢 Kimi K2.6 → done in 3:45, 16.7k output tokens 🟡 Step 3.7 Flash → half the fields, 57k output tokens 🔴 Gemini 3.5 Flash → 489k output tokens, never finished. I had to kill it. Gemini burned ~29x more output tokens than Kimi on the exact same task, and Kimi's was the only form that actually looked filled out. The test, a mocked application form, contains some challenging parts, such as one-character-per-box fields. I provided every model the same set of tools: > get canvas size > drop probe markers to find coordinates > add text > add checkmarks > move elements > take a screenshot anytime to check their own work > ... etc So it's vision + spatial reasoning + tool use + long context, all at once. Small models (Qwen, Gemma) can't really complete this test, so I skipped them. What happened: > Kimi nailed name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code — placement slightly loose, but content correct. 15 turns. Clean. > Step got maybe half right — fields dropped, "United States" landed in the email line, data floating outside boxes. Burned 1.24M input tokens doing it (81 turns of re-reading the canvas). > Gemini almost got there visually... then spiraled. By turn 40 it was issuing a delete_elements call wiping element IDs 365–425, basically erasing its own work. 31 minutes, 489k output tokens, still streaming. Terminated. The takeaway isn't "Gemini bad." This test is indeed difficult. But token efficiency is capability now. A model that needs 30x the tokens and still can't converge is going to be 30x the cost in production. Kimi K2.6 just quietly did the thing.
0:38

I designed a new test specifically for multimodal models: fill out a paper form. And it's much harder than it sounds. This isn't typing into an electronic field that captures your text. The form is just an image. The model has to place each form element: text, checkmarks — at the correct pixel position on the canvas itself. Results: 🟢 Kimi K2.6 → done in 3:45, 16.7k output tokens 🟡 Step 3.7 Flash → half the fields, 57k output tokens 🔴 Gemini 3.5 Flash → 489k output tokens, never finished. I had to kill it. Gemini burned ~29x more output tokens than Kimi on the exact same task, and Kimi's was the only form that actually looked filled out. The test, a mocked application form, contains some challenging parts, such as one-character-per-box fields. I provided every model the same set of tools: > get canvas size > drop probe markers to find coordinates > add text > add checkmarks > move elements > take a screenshot anytime to check their own work > ... etc So it's vision + spatial reasoning + tool use + long context, all at once. Small models (Qwen, Gemma) can't really complete this test, so I skipped them. What happened: > Kimi nailed name, DOB, ID, gender, marital status, nationality, email, phone, address, postal code — placement slightly loose, but content correct. 15 turns. Clean. > Step got maybe half right — fields dropped, "United States" landed in the email line, data floating outside boxes. Burned 1.24M input tokens doing it (81 turns of re-reading the canvas). > Gemini almost got there visually... then spiraled. By turn 40 it was issuing a delete_elements call wiping element IDs 365–425, basically erasing its own work. 31 minutes, 489k output tokens, still streaming. Terminated. The takeaway isn't "Gemini bad." This test is indeed difficult. But token efficiency is capability now. A model that needs 30x the tokens and still can't converge is going to be 30x the cost in production. Kimi K2.6 just quietly did the thing.

stevibe

25,455 views • 2 months ago

i watched gemma 4 12b build something genuinely impressive today, and then loop itself to death right in front of me. the full run is in the video, sped up but completely uncut, watch it to the end and you will catch the exact moment it stops building and starts looping right in the middle of the work. the task was clean, build a single file gravity simulator, n-body physics, orbits, collisions, running locally on one 3090 through an agent. and for ten minutes it was a joy to watch. it reached for a symplectic integrator on its own, the correct one, the kind that keeps orbits stable instead of spiralling out. real gravity with softening, proper orbital velocities, momentum conserved on collision. the physics was right. the thing actually worked. then on the very last step, writing a few tests to prove its own code, it fell into a loop. not a crash, a loop. it started repeating itself and would not stop. ten more minutes, thirty four thousand tokens into a single answer, the same fragments over and over, until i killed it myself. so it's not that gemma can't code. it did the hard part beautifully. it cannot finish. it cannot hold a long task together without unravelling, and finishing is the entire job in agentic work. here's the part that stings. i run this exact task, same harness, same card, on the chinese open models, qwen especially, and i never see this. they build it, they test it, they stop. every single time. google has the raw capability, you can see it sitting right there in the code, and then the model loops itself to death on a task a 27b from alibaba finishes clean. open weights, apache 2.0, so much to love on paper. i just need it to know when to stop talking.
4:15

i watched gemma 4 12b build something genuinely impressive today, and then loop itself to death right in front of me. the full run is in the video, sped up but completely uncut, watch it to the end and you will catch the exact moment it stops building and starts looping right in the middle of the work. the task was clean, build a single file gravity simulator, n-body physics, orbits, collisions, running locally on one 3090 through an agent. and for ten minutes it was a joy to watch. it reached for a symplectic integrator on its own, the correct one, the kind that keeps orbits stable instead of spiralling out. real gravity with softening, proper orbital velocities, momentum conserved on collision. the physics was right. the thing actually worked. then on the very last step, writing a few tests to prove its own code, it fell into a loop. not a crash, a loop. it started repeating itself and would not stop. ten more minutes, thirty four thousand tokens into a single answer, the same fragments over and over, until i killed it myself. so it's not that gemma can't code. it did the hard part beautifully. it cannot finish. it cannot hold a long task together without unravelling, and finishing is the entire job in agentic work. here's the part that stings. i run this exact task, same harness, same card, on the chinese open models, qwen especially, and i never see this. they build it, they test it, they stop. every single time. google has the raw capability, you can see it sitting right there in the code, and then the model loops itself to death on a task a 27b from alibaba finishes clean. open weights, apache 2.0, so much to love on paper. i just need it to know when to stop talking.

Sudo su

39,574 views • 2 months ago

qwen 3.8 max vs deepseek v4 flash 0731 vs kimi k3 vs gpt 5.6 sol – on rubik's cube and chess four frontier models built a rubik's cube stand and solved it, then built a chess board and played claude opus 5 on it the setup: Nous Research's hermes agent cli on OpenRouter tasks: 1. cube – build a 3d rubik's cube with a cli and a Three.js viewer, then solve an identical scrambled position on your own stand 2. chess – build a 3d chess stand, then play white against claude opus 5 as black, live, one move at a time. no engine, no solver, no opening book on either side. stockfish depth 14 grades every chess ply afterwards; neither player sees the score models: DeepSeek v4 flash 0731, OpenAI gpt-5.6 sol, Kimi.ai kimi k3, Qwen qwen 3.8 max gpt-5.6 sol and deepseek v4 flash solved their cubes – sol in 24 moves and seventeen seconds, deepseek in 32. qwen and kimi never got there, giving up at 96 and 207 moves then all four built chess stands and played white against claude opus 5 on them, and all four resigned: deepseek on move 13, sol on 19, kimi on 21, qwen holding out longest at 29 - build time, both stands #1 gpt-5.6 sol – 16m 43s #2 deepseek v4 flash – 97m 39s #3 kimi k3 – 166m 09s #4 qwen 3.8 max – 215m 08s - build attempts before a working stand #1 gpt-5.6 sol – 3 #2 qwen 3.8 max – 4 #3 kimi k3 – 4 #4 deepseek v4 flash – 5 - total tokens #1 gpt-5.6 sol – 6,713,754 #2 qwen 3.8 max – 17,272,507 #3 kimi k3 – 22,427,504 #4 deepseek v4 flash – 27,417,442 - total price #1 deepseek v4 flash – $0.557 #2 gpt-5.6 sol – $6.319 #3 qwen 3.8 max – $10.270 #4 kimi k3 – $16.667 observations: • deepseek v4 flash is the cheapest model here by a margin nobody else is near, and it got there while being the least efficient of the four. it burned 27.4m tokens – more than anyone, 5m more than kimi – and still finished both benchmarks for $0.557. that is $0.02 per million tokens against kimi's $0.74. it also needed the most passes to produce working stands, five, and that did not matter: all five deepseek passes together cost a thirtieth of kimi's two • so what deepseek cannot do is get it right the first time. what it can do is get it right the fifth time, for half a dollar. that is a different thing to be buying – not a good first draft, but the option to keep asking • gpt-5.6 sol is the opposite profile and the strongest of the four on pure efficiency. 16m 43s to build both stands, 6.7m tokens, three passes – under 40% of the next lowest token count and a quarter of deepseek's, on an eighth of qwen's clock. it also solved the cube fastest of anyone, 24 moves in seventeen seconds. sol is what you reach for when you want the answer now and can absorb $0.94 per million • sol's weakness is in what it does not check. its chess viewer deleted the capturing piece instead of the captured one, so pieces disappeared off the board mid-game – a defect the fifty-cent deepseek stand did not have. fast and terse turns out to be the same dial as fast and unverified • qwen 3.8 max is not the cheap open-weights option it gets treated as. $10.270 across the two benchmarks, second most expensive of the four, 18x deepseek, and by a distance the slowest – 215 minutes of build time, nearly thirteen times sol's. what the money buys is judgment: it played eighteen moves without a single error worth a hundredth of a pawn, then made exactly one bad move in the whole game, and averaged 44.6 centipawns lost across the longest game any of the four managed. it also could not solve a rubik's cube in 96 tries • kimi k3 is the one line with no reading that flatters it. most expensive at $16.667, last on the cube at 207 moves, last at chess at 478 centipawns lost per move. it is also the model that verified hardest – on the cube it wrote its own integrity check instead of trusting its output. that makes the result worse rather than better: the checking was real, and the reasoning underneath it still was not follow thehype. for 24/7 ai news, analysis and breakdowns
1:03

qwen 3.8 max vs deepseek v4 flash 0731 vs kimi k3 vs gpt 5.6 sol – on rubik's cube and chess four frontier models built a rubik's cube stand and solved it, then built a chess board and played claude opus 5 on it the setup: Nous Research's hermes agent cli on OpenRouter tasks: 1. cube – build a 3d rubik's cube with a cli and a Three.js viewer, then solve an identical scrambled position on your own stand 2. chess – build a 3d chess stand, then play white against claude opus 5 as black, live, one move at a time. no engine, no solver, no opening book on either side. stockfish depth 14 grades every chess ply afterwards; neither player sees the score models: DeepSeek v4 flash 0731, OpenAI gpt-5.6 sol, Kimi.ai kimi k3, Qwen qwen 3.8 max gpt-5.6 sol and deepseek v4 flash solved their cubes – sol in 24 moves and seventeen seconds, deepseek in 32. qwen and kimi never got there, giving up at 96 and 207 moves then all four built chess stands and played white against claude opus 5 on them, and all four resigned: deepseek on move 13, sol on 19, kimi on 21, qwen holding out longest at 29 - build time, both stands #1 gpt-5.6 sol – 16m 43s #2 deepseek v4 flash – 97m 39s #3 kimi k3 – 166m 09s #4 qwen 3.8 max – 215m 08s - build attempts before a working stand #1 gpt-5.6 sol – 3 #2 qwen 3.8 max – 4 #3 kimi k3 – 4 #4 deepseek v4 flash – 5 - total tokens #1 gpt-5.6 sol – 6,713,754 #2 qwen 3.8 max – 17,272,507 #3 kimi k3 – 22,427,504 #4 deepseek v4 flash – 27,417,442 - total price #1 deepseek v4 flash – $0.557 #2 gpt-5.6 sol – $6.319 #3 qwen 3.8 max – $10.270 #4 kimi k3 – $16.667 observations: • deepseek v4 flash is the cheapest model here by a margin nobody else is near, and it got there while being the least efficient of the four. it burned 27.4m tokens – more than anyone, 5m more than kimi – and still finished both benchmarks for $0.557. that is $0.02 per million tokens against kimi's $0.74. it also needed the most passes to produce working stands, five, and that did not matter: all five deepseek passes together cost a thirtieth of kimi's two • so what deepseek cannot do is get it right the first time. what it can do is get it right the fifth time, for half a dollar. that is a different thing to be buying – not a good first draft, but the option to keep asking • gpt-5.6 sol is the opposite profile and the strongest of the four on pure efficiency. 16m 43s to build both stands, 6.7m tokens, three passes – under 40% of the next lowest token count and a quarter of deepseek's, on an eighth of qwen's clock. it also solved the cube fastest of anyone, 24 moves in seventeen seconds. sol is what you reach for when you want the answer now and can absorb $0.94 per million • sol's weakness is in what it does not check. its chess viewer deleted the capturing piece instead of the captured one, so pieces disappeared off the board mid-game – a defect the fifty-cent deepseek stand did not have. fast and terse turns out to be the same dial as fast and unverified • qwen 3.8 max is not the cheap open-weights option it gets treated as. $10.270 across the two benchmarks, second most expensive of the four, 18x deepseek, and by a distance the slowest – 215 minutes of build time, nearly thirteen times sol's. what the money buys is judgment: it played eighteen moves without a single error worth a hundredth of a pawn, then made exactly one bad move in the whole game, and averaged 44.6 centipawns lost across the longest game any of the four managed. it also could not solve a rubik's cube in 96 tries • kimi k3 is the one line with no reading that flatters it. most expensive at $16.667, last on the cube at 207 moves, last at chess at 478 centipawns lost per move. it is also the model that verified hardest – on the cube it wrote its own integrity check instead of trusting its output. that makes the result worse rather than better: the checking was real, and the reasoning underneath it still was not follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

80,750 views • 7 days ago

Which LLMs actually love to think? Tested 7 models on 5 math problems, measured reasoning length. The think winners: both Qwen3.5 models (27B and 35B A3B) — massive overthinkers, up to 10k+ tokens on a single question. Plot twists: > Kimi K2.6 feels verbose, actually one of the leanest > Gemma4 26B A4B solved 2 with ZERO thinking
0:42

Which LLMs actually love to think? Tested 7 models on 5 math problems, measured reasoning length. The think winners: both Qwen3.5 models (27B and 35B A3B) — massive overthinkers, up to 10k+ tokens on a single question. Plot twists: > Kimi K2.6 feels verbose, actually one of the leanest > Gemma4 26B A4B solved 2 with ZERO thinking

stevibe

97,403 views • 3 months ago

Six open-source LLMs. One sliding puzzle. A brutal test of long-horizon reasoning and tool calling. Five of them broke. One didn't. I gave each model a move_tile tool and a scrambled 3×3 board, then asked it to solve the puzzle through pure turn-by-turn reasoning. The deeper the scramble, the more brutal the search. Five runs per depth, best run kept. A model fails the round if it exceeds 6x the optimal move count. > Depth 5: Everyone solves it. Yawn. > Depth 10: GLM 5.1 melts down. 43 moves. Cut. > Depth 12: Gemma4 26B loses the plot, shuffling tiles in circles. Gone. > Depth 15: The wall. DeepSeek V4 Flash, out. DeepSeek V4 Pro, out. Gemma4, out again. GLM 5.1, out. Two survivors: Qwen3.6 35B-A3B, and Kimi K2.6 with an 11-move solve that looked like cheating. > Depth 18: Same two. Everyone else hallucinating tiles that weren't there. > Depth 22: Final boss. Kimi, flawless for five rounds, finally cracks. 81 moves. Still scrambled. DeepSeek V4 Pro limps home at 90. Qwen3.6 35B-A3B solves it in 36. The smallest model in the room. ~3B active params. Fits on a single 3090. It beat everything. Kimi was elegant. Qwen3.6 was unstoppable.
2:00

Six open-source LLMs. One sliding puzzle. A brutal test of long-horizon reasoning and tool calling. Five of them broke. One didn't. I gave each model a move_tile tool and a scrambled 3×3 board, then asked it to solve the puzzle through pure turn-by-turn reasoning. The deeper the scramble, the more brutal the search. Five runs per depth, best run kept. A model fails the round if it exceeds 6x the optimal move count. > Depth 5: Everyone solves it. Yawn. > Depth 10: GLM 5.1 melts down. 43 moves. Cut. > Depth 12: Gemma4 26B loses the plot, shuffling tiles in circles. Gone. > Depth 15: The wall. DeepSeek V4 Flash, out. DeepSeek V4 Pro, out. Gemma4, out again. GLM 5.1, out. Two survivors: Qwen3.6 35B-A3B, and Kimi K2.6 with an 11-move solve that looked like cheating. > Depth 18: Same two. Everyone else hallucinating tiles that weren't there. > Depth 22: Final boss. Kimi, flawless for five rounds, finally cracks. 81 moves. Still scrambled. DeepSeek V4 Pro limps home at 90. Qwen3.6 35B-A3B solves it in 36. The smallest model in the room. ~3B active params. Fits on a single 3090. It beat everything. Kimi was elegant. Qwen3.6 was unstoppable.

stevibe

16,434 views • 3 months ago

"Its (Sonnet 5) performance is close to Opus 4.8, at lower prices." So I ran 4 canvas test through both. > Opus 4.8, 4/4 actually animating. > Sonnet 5, 2/4 came back as static images. And "lower price"? On the paper shredder task, Sonnet 5 spent $0.36 for a static image. Opus 4.8 spent $0.18 and it actually animated. The 4 tests: > Win 98 drag-to-BSOD > Self-typing keyboard + CRT > Letter burning > Paper shredder
0:34

"Its (Sonnet 5) performance is close to Opus 4.8, at lower prices." So I ran 4 canvas test through both. > Opus 4.8, 4/4 actually animating. > Sonnet 5, 2/4 came back as static images. And "lower price"? On the paper shredder task, Sonnet 5 spent $0.36 for a static image. Opus 4.8 spent $0.18 and it actually animated. The 4 tests: > Win 98 drag-to-BSOD > Self-typing keyboard + CRT > Letter burning > Paper shredder

stevibe

60,788 views • 1 month ago

hear this anon you don't need a $4,699 box to get started local AI. use what you already have first. test your workload. this is what a $250 GPU did today. iteration 3 of octopus invaders is here. 4 phases. 6 prompts. zero handwritten code. the same 9B on the same 3060 fixed its own enemy spawning, patched a dual start conflict, added level progression, resized every bullet, and when the browser cached old files it figured that out on its own and added version parameters to force reload. 3,200+ lines across 13 files. every line by qwen 3.5 9B Q4 at 35-50 tok/s on 12 gigs through hermes agent. understand what your load actually needs before you build. don't get trapped by influencers selling you boxes next to a plant. test on what you have. then decide. this 3060 impressed me in ways i did not expect and its autonomy is what kept me going. now its time to move to new experiments on other nodes and other models for all of us. if you are running this setup the exact stack, flags, and open source code, exact prompts i used are in the replies. if you run into issues let me know. seeing students and builders discover hermes from my posts and start running local is why i do this. full autonomous build at 8x speed in the video. gameplay at the end. watch it.
5:51

hear this anon you don't need a $4,699 box to get started local AI. use what you already have first. test your workload. this is what a $250 GPU did today. iteration 3 of octopus invaders is here. 4 phases. 6 prompts. zero handwritten code. the same 9B on the same 3060 fixed its own enemy spawning, patched a dual start conflict, added level progression, resized every bullet, and when the browser cached old files it figured that out on its own and added version parameters to force reload. 3,200+ lines across 13 files. every line by qwen 3.5 9B Q4 at 35-50 tok/s on 12 gigs through hermes agent. understand what your load actually needs before you build. don't get trapped by influencers selling you boxes next to a plant. test on what you have. then decide. this 3060 impressed me in ways i did not expect and its autonomy is what kept me going. now its time to move to new experiments on other nodes and other models for all of us. if you are running this setup the exact stack, flags, and open source code, exact prompts i used are in the replies. if you run into issues let me know. seeing students and builders discover hermes from my posts and start running local is why i do this. full autonomous build at 8x speed in the video. gameplay at the end. watch it.

Sudo su

51,804 views • 4 months ago

2.3x faster. Ran Unsloth AI Qwen3.6 MTP variants on a DGX Spark (UD-Q6_K_XL): &gt; 27B → 27B MTP: 8.1 → 18.65 t/s (2.3x faster) &gt; 35B A3B → 35B A3B MTP: 56.91 → 66.52 t/s (+17%) The 27B dense model more than doubled throughput from MTP alone. Free speed is free speed.
1:00

2.3x faster. Ran Unsloth AI Qwen3.6 MTP variants on a DGX Spark (UD-Q6_K_XL): &gt; 27B → 27B MTP: 8.1 → 18.65 t/s (2.3x faster) &gt; 35B A3B → 35B A3B MTP: 56.91 → 66.52 t/s (+17%) The 27B dense model more than doubled throughput from MTP alone. Free speed is free speed.

stevibe

24,315 views • 3 months ago

Qwen3.5:9b reasoning head-to-head: Mac Studio M2 Ultra 64GB: 43.08 tok/s Mac Mini M4 16GB: 13.07 tok/s Qwen
1:18

Qwen3.5:9b reasoning head-to-head: Mac Studio M2 Ultra 64GB: 43.08 tok/s Mac Mini M4 16GB: 13.07 tok/s Qwen

stevibe

243,343 views • 5 months ago

Qwen3.6 35B-A3B dropped yesterday, so I ran it on 4 GPUs to see how it performs: 🟣 RTX 3090 — 49.78 tok/s, TTFT 852ms 🟡 RTX 4090 — 118.93 tok/s, TTFT 686ms 🟢 RTX 5090 — 160.37 tok/s, TTFT 409ms 🔵 DGX Spark — 59.98 tok/s, TTFT 228ms I went with ollama as the backend because honestly, it's the easiest way for most people to get started. One command, model pulled, done. I used Q4_K_M (24GB) across all four cards. The reason is the 3090 and 4090 don't support NVFP4 (only the 5090 and DGX Spark could use it). Keeping the same quant everywhere felt like the fairest way to compare. And yes, you can absolutely squeeze more performance out of every card with vLLM, SGLang, or TensorRT-LLM. But that's not what this test is about. This is just the out-of-the-box experience for folks who own a GPU and want to try the new model tonight.
1:24

Qwen3.6 35B-A3B dropped yesterday, so I ran it on 4 GPUs to see how it performs: 🟣 RTX 3090 — 49.78 tok/s, TTFT 852ms 🟡 RTX 4090 — 118.93 tok/s, TTFT 686ms 🟢 RTX 5090 — 160.37 tok/s, TTFT 409ms 🔵 DGX Spark — 59.98 tok/s, TTFT 228ms I went with ollama as the backend because honestly, it's the easiest way for most people to get started. One command, model pulled, done. I used Q4_K_M (24GB) across all four cards. The reason is the 3090 and 4090 don't support NVFP4 (only the 5090 and DGX Spark could use it). Keeping the same quant everywhere felt like the fairest way to compare. And yes, you can absolutely squeeze more performance out of every card with vLLM, SGLang, or TensorRT-LLM. But that's not what this test is about. This is just the out-of-the-box experience for folks who own a GPU and want to try the new model tonight.

stevibe

401,804 views • 3 months ago