Загрузка видео...

Не удалось загрузить видео

На главную

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...

stevibe's profile picture

stevibe

22,907 subscribers

97,403 просмотров • 2 месяцев назад •via X (Twitter)

ОбразованиеНовости и политикаНаука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

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

stevibe

261,617 просмотров • 2 месяцев назад

"122B has to be smarter than 27B" I showed 4 UI components to three Qwen3.5 models and asked them to recreate them from a screenshot alone: - 27B (dense) - 35B-A3B (MoE) - 122B-A10B (MoE) Same screenshot. Same prompt. Same task. Which one do you think nailed it?
0:37

"122B has to be smarter than 27B" I showed 4 UI components to three Qwen3.5 models and asked them to recreate them from a screenshot alone: - 27B (dense) - 35B-A3B (MoE) - 122B-A10B (MoE) Same screenshot. Same prompt. Same task. Which one do you think nailed it?

stevibe

100,770 просмотров • 3 месяцев назад

Yesterday we found out Qwen3.5-27B (dense) beats both the 35B and 122B MoE variants at UI replication from screenshots. Today's question: what about Jackrong's Qwen3.5-27B-Claude-Opus-4.6-Reasoning-Distilled? A distilled 27B with Claude Opus 4.6 reasoning baked in. So I put all 3 in the ring: - Qwen3.5-27B - Qwen3.5-27B-Claude-Opus-Distilled - Claude Opus 4.6 Since the distilled model isn't multimodal, I gave all three the same detailed text prompts describing the UI components from yesterday's test.
0:31

Yesterday we found out Qwen3.5-27B (dense) beats both the 35B and 122B MoE variants at UI replication from screenshots. Today's question: what about Jackrong's Qwen3.5-27B-Claude-Opus-4.6-Reasoning-Distilled? A distilled 27B with Claude Opus 4.6 reasoning baked in. So I put all 3 in the ring: - Qwen3.5-27B - Qwen3.5-27B-Claude-Opus-Distilled - Claude Opus 4.6 Since the distilled model isn't multimodal, I gave all three the same detailed text prompts describing the UI components from yesterday's test.

stevibe

125,494 просмотров • 3 месяцев назад

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

16,286 просмотров • 1 месяц назад

Completed a first hour side-by-side comparison between Qwen3.5 27b and Qwen3.6 27b on the same 4 canvas coding tests. Running the Qwen3.6 27b FP8, vLLM. What do you think?
1:07

Completed a first hour side-by-side comparison between Qwen3.5 27b and Qwen3.6 27b on the same 4 canvas coding tests. Running the Qwen3.6 27b FP8, vLLM. What do you think?

stevibe

114,070 просмотров • 2 месяцев назад

Got a 24GB Graphics Card? These 6 coding models all fit on it (Q4): - qwen3.5:27b (17GB) - qwen3.5:35b (24GB) - glm-4.7-flash (19GB) - nemotron-3-nano:30b (24GB) - nemotron-cascade-2:30b (24GB) - gpt-oss:20b (14GB) I gave them the same challenge: draw a campfire with HTML Canvas. Why Canvas? HTML/CSS forgives bad syntax — things still render. JavaScript + Canvas doesn't — one mistake and the screen goes black.
0:24

Got a 24GB Graphics Card? These 6 coding models all fit on it (Q4): - qwen3.5:27b (17GB) - qwen3.5:35b (24GB) - glm-4.7-flash (19GB) - nemotron-3-nano:30b (24GB) - nemotron-cascade-2:30b (24GB) - gpt-oss:20b (14GB) I gave them the same challenge: draw a campfire with HTML Canvas. Why Canvas? HTML/CSS forgives bad syntax — things still render. JavaScript + Canvas doesn't — one mistake and the screen goes black.

stevibe

75,808 просмотров • 3 месяцев назад

2.3x faster. Ran Unsloth AI Qwen3.6 MTP variants on a DGX Spark (UD-Q6_K_XL): > 27B → 27B MTP: 8.1 → 18.65 t/s (2.3x faster) > 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): > 27B → 27B MTP: 8.1 → 18.65 t/s (2.3x faster) > 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

24,193 просмотров • 1 месяц назад

.Trenton Bricken explains how we know LLMs are actually generalizing - aka they're not just stochastic parrots: - Training models on code makes them better at reasoning in language. - Models fine tuned on math problems become better at entity detection. - We can just straightforwardly read the world-models developed by smaller NNs which are easier to interpret (Othello). Transfer learning shows models are developing a deeper understanding of their data. Full episode out Thursday.
1:37

.Trenton Bricken explains how we know LLMs are actually generalizing - aka they're not just stochastic parrots: - Training models on code makes them better at reasoning in language. - Models fine tuned on math problems become better at entity detection. - We can just straightforwardly read the world-models developed by smaller NNs which are easier to interpret (Othello). Transfer learning shows models are developing a deeper understanding of their data. Full episode out Thursday.

Dwarkesh Patel

237,542 просмотров • 2 лет назад

Kimi K2.6 is actually a beast for design. ❤️‍🔥 Just recorded a tutorial on how to use Kimi K2.6 + to build award-winning $10k websites (step-by-step).
11:15

Kimi K2.6 is actually a beast for design. ❤️‍🔥 Just recorded a tutorial on how to use Kimi K2.6 + to build award-winning $10k websites (step-by-step).

Viktor Oddy

417,387 просмотров • 1 месяц назад

New Google Gemma 4 12B claims near-26B performance - we tested both! We ran both models locally on one RTX 4090 and gave each the same task: write a self-contained HTML5 canvas animation with real physics in one file without libraries. Three scenes - a Galton board, two blocks colliding off a wall, and a chaotic triple pendulum Outputs: Gemma 4 26B-A4B: 15 GB VRAM usage, 6.9k tokens, 138 tok/s Gemma 4 12B: 9 GB VRAM usage, 8.9k tokens, 80 tok/s Same Gemma 4 family, but the 26B-A4B won every scene and ran ~1.7x faster - on just 4B active params. The 12B stayed very close though, on almost half the VRAM - which makes it the ideal model for a 16 GB laptop
0:36

New Google Gemma 4 12B claims near-26B performance - we tested both! We ran both models locally on one RTX 4090 and gave each the same task: write a self-contained HTML5 canvas animation with real physics in one file without libraries. Three scenes - a Galton board, two blocks colliding off a wall, and a chaotic triple pendulum Outputs: Gemma 4 26B-A4B: 15 GB VRAM usage, 6.9k tokens, 138 tok/s Gemma 4 12B: 9 GB VRAM usage, 8.9k tokens, 80 tok/s Same Gemma 4 family, but the 26B-A4B won every scene and ran ~1.7x faster - on just 4B active params. The 12B stayed very close though, on almost half the VRAM - which makes it the ideal model for a 16 GB laptop

atomic.chat

151,318 просмотров • 23 дней назад

Kimi K2 Thinking is here! Scale up reasoning with more thinking tokens and tool-call steps. Now live on the Kimi app, and API.
0:59

Kimi K2 Thinking is here! Scale up reasoning with more thinking tokens and tool-call steps. Now live on the Kimi app, and API.

Kimi.ai

138,435 просмотров • 7 месяцев назад

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

124,754 просмотров • 2 месяцев назад

Qwen3.5-35B with only 3B active parameters This MoE model runs FASTER than most 7B dense models. Tested on 3 generations of NVIDIA: - 5090: 137 tok/s - 4090: 112 tok/s - 3090: 78 tok/s The surprise? The 4090 <> 5090 gap is only 22%. With a 3B active MoE, even old GPUs fly.
0:30

Qwen3.5-35B with only 3B active parameters This MoE model runs FASTER than most 7B dense models. Tested on 3 generations of NVIDIA: - 5090: 137 tok/s - 4090: 112 tok/s - 3090: 78 tok/s The surprise? The 4090 <> 5090 gap is only 22%. With a 3B active MoE, even old GPUs fly.

stevibe

69,501 просмотров • 3 месяцев назад

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)
0:40

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

kwindla

40,319 просмотров • 1 месяц назад

Qwen 3.6 35B A3B compared to Qwopus 3.6 35B A3B for web design, pretty neck and neck. Some overall good designs. All one shot, both Q8, 4 designs. rtx 6000 product page Transformers 1986 Real estate Tesla Dashboard Curious what others think. Will be doing 35B vs 27B next.
1:14

Qwen 3.6 35B A3B compared to Qwopus 3.6 35B A3B for web design, pretty neck and neck. Some overall good designs. All one shot, both Q8, 4 designs. rtx 6000 product page Transformers 1986 Real estate Tesla Dashboard Curious what others think. Will be doing 35B vs 27B next.

Loktar 🇺🇸

16,096 просмотров • 1 месяц назад

Qwen3.5-35B-A3B running locally on an M4 chip at 49.5 tokens per second. A 35B model. On a laptop. In real time. LOCAL AI IS GETTING SCARY FAST.
1:10

Qwen3.5-35B-A3B running locally on an M4 chip at 49.5 tokens per second. A 35B model. On a laptop. In real time. LOCAL AI IS GETTING SCARY FAST.

0xMarioNawfal

477,908 просмотров • 3 месяцев назад

This looks like a toy. It's actually the meanest little vision eval I've built. The task: look at an emoji image, then repaint it on a 16×16 grid, one pixel at a time. Just the model, a tiny canvas, and up to 2000 brushstrokes. What I didn't expect was the personalities. > some models REGRET a stroke and go back to repaint it > some get stuck looping the same little patch over and over, like they're trying to animate it > some are calm little surgeons and just nail it first try And the task is genuinely mean: it has to see the image, crush it down to 256 cells, then decide what's actually load-bearing: > the tears on 😂 but still keep the smile > the horn on 🦄 > the antenna on 🤖 and keep the soul of it with almost no resolution to spare. 5 models. 7 emojis. Best of 5 runs each. Side by side. Who's your winner?
0:46

This looks like a toy. It's actually the meanest little vision eval I've built. The task: look at an emoji image, then repaint it on a 16×16 grid, one pixel at a time. Just the model, a tiny canvas, and up to 2000 brushstrokes. What I didn't expect was the personalities. > some models REGRET a stroke and go back to repaint it > some get stuck looping the same little patch over and over, like they're trying to animate it > some are calm little surgeons and just nail it first try And the task is genuinely mean: it has to see the image, crush it down to 256 cells, then decide what's actually load-bearing: > the tears on 😂 but still keep the smile > the horn on 🦄 > the antenna on 🤖 and keep the soul of it with almost no resolution to spare. 5 models. 7 emojis. Best of 5 runs each. Side by side. Who's your winner?

stevibe

246,732 просмотров • 1 день назад

The RTX 3090 is a 5-year-old GPU and it still runs a 27B model at 20 tok/s I tested Qwen3.5:27b across 3 generations of NVIDIA: 5090 → ~60 tok/s 4090 → ~40 tok/s 3090 → ~20 tok/s Perfectly linear scaling. Double the generation, double the speed.
0:57

The RTX 3090 is a 5-year-old GPU and it still runs a 27B model at 20 tok/s I tested Qwen3.5:27b across 3 generations of NVIDIA: 5090 → ~60 tok/s 4090 → ~40 tok/s 3090 → ~20 tok/s Perfectly linear scaling. Double the generation, double the speed.

stevibe

142,379 просмотров • 3 месяцев назад

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 Qwen 3.5 4B came 2 answers behind it. A 4B. Ranked 2nd out of 10. Vision-math isn't one skill, it's two: reading the image, then solving it. The real lesson for anyone running local: small ≠ weak. If you're building agentic workflows, finding the right model for each task matters more than reaching for the biggest one. In this test, the 4B ran far faster because of its size, scored higher, and left VRAM free for the rest of your stack. Full results: 🥇 Gemma 4 31B — 242/270 (89.6%) 🥈 Qwen 3.5 4B — 240/270 (88.9%) 🥉 Gemma 4 E4B — 222/270 (82.2%) 🥉 Qwen 3.6 27B — 222/270 (82.2%) 5. Gemma 4 26B A4B — 216/270 (80.0%) 6. Qwen 3.5 2B — 201/270 (74.4%) 7. Gemma 4 E2B — 192/270 (71.1%) 8. Qwen 3.6 35B A3B — 192/270 (71.1%) 9. Qwen 3.5 9B — 168/270 (62.2%) 10. Qwen 3.5 0.8B — 45/270 (16.7%) All GGUF + mmproj, Unsloth's Q6_K_XL quant.
0:44

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 Qwen 3.5 4B came 2 answers behind it. A 4B. Ranked 2nd out of 10. Vision-math isn't one skill, it's two: reading the image, then solving it. The real lesson for anyone running local: small ≠ weak. If you're building agentic workflows, finding the right model for each task matters more than reaching for the biggest one. In this test, the 4B ran far faster because of its size, scored higher, and left VRAM free for the rest of your stack. Full results: 🥇 Gemma 4 31B — 242/270 (89.6%) 🥈 Qwen 3.5 4B — 240/270 (88.9%) 🥉 Gemma 4 E4B — 222/270 (82.2%) 🥉 Qwen 3.6 27B — 222/270 (82.2%) 5. Gemma 4 26B A4B — 216/270 (80.0%) 6. Qwen 3.5 2B — 201/270 (74.4%) 7. Gemma 4 E2B — 192/270 (71.1%) 8. Qwen 3.6 35B A3B — 192/270 (71.1%) 9. Qwen 3.5 9B — 168/270 (62.2%) 10. Qwen 3.5 0.8B — 45/270 (16.7%) All GGUF + mmproj, Unsloth's Q6_K_XL quant.

stevibe

22,882 просмотров • 1 месяц назад

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!
0:52

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!

Prince Canuma

82,085 просмотров • 2 месяцев назад