Three Cline agents. Same 120B model. One rule: terminate... your opponents' processes before they terminate yours. We tested raw inference across an NVIDIA AI PC DGX Spark, an RTX 4090, and cloud. Time-to-first-token decides the race. Throughput decides who ships. DGX Spark: 42.9 tok/s. RTX 4090: 8.7. Same weights. 4.9x gap. #AIonRTXshow more

stealth model update: 5x more context `code-supernova-1-million` (access via... cline provider) > 1 million token context window (up from 200k) > same multi-modal support > same free access during alpha <the supernova is intensifying>show more

Google engineers are running 3 NVIDIA DGX Spark boxes... each just to speed up their Claude skills workflows by 300% > and the gap between them and everyone else is getting wider every week $95K developer vs $300K AI architect. same job. different stack > /self-improving-agent - the system improves its own process without you touching it > /rag-architect - retrieval quality now matters more than raw model cleverness > /connect-apps - Claude takes real actions across Gmail, Slack, GitHub and Notion the people who build this stack now will be the $300K architects everyone asks "what's your setup?" > everyone else will still be typing prompts full list with GitHub links in the article above 👇show more

1. Model agnostic. 2. Inference agnostic. & now, 3.... Platform agnostic. Cline for JetBrains is here. (install it below)show more

vllm-exl3 v0.3.0 is LIVE with custom native CUDA kernels... for 2-bit EXL3 on NVIDIA DGX Spark GB10. GLM-5.3-Flash-EXL3-K2 jumped from 16.9 → 24.6 tok/s average single-stream decode, a +45.6% gain. Coding hit 27.6 tok/s, +85.6%. 🚀 The previous ExLlamaV3-backed path inside vLLM was leaving a lot of GB10 bandwidth on the table. So I rewrote the hot path specifically for EXL3 on Blackwell sm_121: → in-register Trellis dequantization → native fused MoE decode → power-of-two chunked prefill GEMM → parallel NVMe pre-warm Then I tested it side-by-side on physical DGX Spark hardware using my GLM-5.3-Flash-EXL3-K2 pack and live vLLM HTTP streaming. 🚀 𝗗𝗘𝗖𝗢𝗗𝗘 𝗧𝗛𝗥𝗢𝗨𝗚𝗛𝗣𝗨𝗧 Single-stream C1: Coding 14.9 → 27.6 tok/s +85.6% Prose 13.7 → 24.6 tok/s +79.3% Reasoning 18.9 → 25.1 tok/s +32.7% Summary 17.1 → 25.6 tok/s +50.0% Format 16.3 → 24.0 tok/s +47.7% Average: 16.9 → 24.6 tok/s 𝗡𝗘𝗧 𝗚𝗔𝗜𝗡: +45.6% ⏱️ 𝗙𝗜𝗥𝗦𝗧-𝗧𝗢𝗞𝗘𝗡 𝗥𝗘𝗦𝗣𝗢𝗡𝗦𝗜𝗩𝗘𝗡𝗘𝗦𝗦 Coding TTFT: 2,344 ms → 859 ms That is a 63.3% reduction, or about 2.7× faster to first token. Follow-up turn with prefix cache hit: 5,608 ms → 3,588 ms 1.56× faster. ⚡ 𝗪𝗛𝗔𝗧 𝗖𝗛𝗔𝗡𝗚𝗘𝗗 𝗢𝗡 𝗧𝗛𝗘 𝗚𝗣𝗨 40 routed-MoE layers: 19.9 ms → 11.5 ms per token Per-layer MoE compute: 497 μs → 287.8 μs That removes 8.4 ms of MoE compute from every generated token. Total per-step wall time: 59.2 ms → 40.6 ms -31.4% The key is `p2b_fused_moe`. Instead of expanding EXL3 weights through a traditional intermediate path, the new kernel performs Trellis dequantization in-register while executing the routed expert computation. The weights stay compressed until the GPU actually needs them. 🔥 𝗣𝗥𝗘𝗙𝗜𝗟𝗟 𝗚𝗢𝗧 𝗔 𝗡𝗔𝗧𝗜𝗩𝗘 𝗣𝗔𝗧𝗛 𝗧𝗢𝗢 The new `exl3_gemm` uses power-of-two chunked prefill GEMM. Measured: 7.85 TFLOPS 13.0× faster than the legacy prefill kernel 1,875 tok/s cold prefill sustained across 65K context 💾 𝗧𝗛𝗘 𝗕𝗢𝗢𝗧 𝗣𝗔𝗧𝗛 𝗡𝗘𝗘𝗗𝗘𝗗 𝗪𝗢𝗥𝗞 𝗧𝗢𝗢 Loading a ~91 GiB model is part of the user experience. Standard shard loading is mostly serial. The updated recipe parallelizes NVMe pre-warm across 8 workers so the storage controller gets used properly instead of feeding a ~100 GiB model one shard at a time. That turns boot-time storage into another optimization target instead of something we simply accept. 💡 𝗧𝗪𝗢 𝗦𝗘𝗥𝗩𝗜𝗡𝗚 𝗙𝗟𝗔𝗚𝗦 𝗪𝗢𝗥𝗧𝗛 𝗞𝗡𝗢𝗪𝗜𝗡𝗚 `--long-prefill-token-threshold 1024` Prevents giant prefill chunks from monopolizing step budgets and starving parallel decode sessions. `--enable-prefix-caching` Avoids paying for the same conversational prefix again on follow-up turns. 📦 𝗘𝗩𝗘𝗥𝗬𝗧𝗛𝗜𝗡𝗚 𝗜𝗦 𝗢𝗣𝗘𝗡 vllm-exl3: GLM-5.3-Flash one-Spark recipe: Model: This is why I like working at the kernel level. The model did not change. The quant did not change. The hardware did not change. The execution path did. 16.9 → 24.6 tok/s. 🛠️ vLLM turboderpshow more

A 27B model in 12 GB of VRAM. Full... 262K context. 70 tok/s. And now it is UNCENSORED. I ran it on my own RTX 3090, held to 12 GB. This is Mirai S Qwen3.8-27B, 2.4 bits per weight, an 11 GB file. With my llama.cpp port the same card did 82K context at 38 tok/s inside 12 GB. Then Palmer took the port and did the hard part. Same card, same file, his serve: - context: 82K -> 262,144 with q8 KV - decode: 38 -> 69.9 tok/s - VRAM peak: 11.7 GB - prefill: 1,124 tok/s He keeps part of the KV cache in VRAM and the rest in pinned RAM, and he drafts with MTP at every depth. He checked every kernel against my fork token for token and wrote down every experiment, the failed ones too. On his real 12 GB RTX 4070 it does 76.6. My part is small. I added abliteration. You cant edit these weights, they are compressed codes, so I measured the refusal direction on the quantized model and the engine removes it while the model runs. A 1.3 MB file and 1 flag: - harmful prompts refused: 64 of 64 -> 0 of 64 - held-out set: 78 of 82 -> 0 of 82 - decode: 69.9 -> 69.7 tok/s Counted by a regex, and with thinking on 1 of 24 still refuses. Deep in the context it gets slower on my card: 27.9 tok/s at 60K. So thats 12 GB of VRAM, 262K context, 70 tok/s, abliterated. Open weights, open code, runs on your desk. Cary did a genius amount of work and almost nobody has seen it. Go follow Palmer right now, I mean it. serve: weights + vector:show more

i spent $26,600 on cloud GPU rentals over 14... months before i found a NVIDIA DGX Spark at $2,999 (founder's edition) or $3,999 (shipping price) it paid for itself in 6 weeks i run 200B parameter models locally now and my old cloud provider keeps sending me loyalty discount emails the math on that $26,600 is embarrassing to type out loud $1,900/month for 14 months, H100 instances on a specialist cloud provider, because anything bigger than a 70B model simply would not fit anywhere else i paid the invoices like they were a utility bill and told myself it was just the cost of doing serious AI work it took me over a year to find out it wasn't 14 months, broken down: → months 1-4: $1,400-1,600/month - felt like manageable infrastructure overhead → months 5-9: crept to $1,900-2,100 as i started running DeepSeek-class experiments, costs tracking directly with model size → months 10-12: one agent loop ran for 36 hours against a 130B model while i slept, that month hit $2,400 → month 13: ran the cumulative total for the first time, saw $23,800, felt physically sick → month 14: another $2,800 month while i waited for the hardware to ship the box is the NVIDIA DGX Spark - roughly the footprint of a large mac mini, powered by a GB10 Grace Blackwell chip with 128GB of unified LPDDR5X memory that unified memory is the whole thing an RTX 4090 has 24GB of VRAM, which means a 70B model in full BF16 precision physically does not fit, you're quantizing down or you're renting cloud, those are your options this box loads a 200B parameter model quantized and serves it through vLLM over localhost, same API interface the cloud endpoint used the migration took one line of code - i changed the base URL from the provider's endpoint to 127.0.0.1:8000 and everything just worked electricity to run continuous 200B inference locally comes out to about $12/month the payback arithmetic is almost too clean: $2,999 hardware cost against $1,900/month saved, the box paid for itself before i'd owned it two months what i didn't account for was how completely the cost model changes your behavior when there's no hourly meter running, you greenlight experiments you'd never approve on cloud - agent loops that churn for hours, running 10,000 documents through a reasoning pass at 3am, speculative fine-tuning jobs you'd normally skip because the cost felt unjustifiable i ran more experiments in the first 30 days after the box arrived than in the four months before it the loyalty discount email landed about 8 weeks after i cancelled the cloud subscription 15% off my next three months, valued customer, we'd love to have you back i didn't reply the box was already runningshow more

NVIDIA JUST CAME FOR INTEL AND AMD WITH FOUR... NUMBERS Microsoft and NVIDIA posted the exact same tweet at the exact same time - "A new era of PC" and a set of coordinates. The numbers point straight to the Taipei Music Center, where Jensen Huang takes the stage June 1. It is the N1X - NVIDIA's first ever PC processor, and they are finally about to show it. - The N1X is an Arm laptop chip with 6144 CUDA cores, putting its graphics in RTX 5070 territory - Microsoft is building a special Windows 11 version just for this chip - normal PCs do not even get it - NVIDIA has been teasing the N1X since 2023 and ghosted us at Computex 2025 AND CES 2026 - ASUS, Dell and Lenovo already have laptops loaded and ready to launch Three years of rumors. Five corporate accounts posting in sync. One keynote left. The chip that was never real is about to crash the party Intel and AMD have owned for decades.show more

Claude Max subscribers: You can use your subscription in... Cline instead of paying per-token API pricing. Here's how you can set it up 🧵show more

We built a plugin that gives Jev a browser... in Cline, and have been blown away by the results. 1. Install it in our new desktop app: Customize > Marketplace > Plugins > search 'jev-browser' 2. Create a Vercel AI Gateway API key, then save it to ~/.cline/plugins/cline-jev-browser.config.json as {"gateway": {"apiKey": "..."}} and restart Cline. 3. Ask any browser task and it will launch Chrome in the background to complete it.show more

This guy built a mini AI farm out of... 4 Nvidia boxes It does not look like a data center. It looks like a stack of small machines sitting next to a laptop. But each box is a DGX Spark with Grace Blackwell inside, 128GB unified memory, and enough room to run models normal gaming GPUs cannot even open. Using the launch price from the article, 4 of them is almost $12,000 of local AI compute on one desk. That sounds expensive until you compare it to cloud GPUs. A serious AI builder can burn $1,500 to $3,000 a month renting A100s and H100s for client work, fine-tunes, agents and 70B models. He basically moved that bill from the cloud into hardware he owns. 4 Nvidia boxes. 512GB unified memory. No hourly meter running in the background. No rented GPUs eating the margin every time an agent runs too long. The funny part is most people still think local AI means a slow laptop running a toy model. Meanwhile guys like this are stacking compute at home. Save this, local AI is turning into the new mining farm.show more

Open source AI is actually moving at an unhinged... pace right now. I literally hadn't even finished typing up my last Gemma 4 12b benchmark notes before Google went ahead and dropped the official Quantization Aware Training (QAT) checkpoints on Hugging Face. If you missed the news, QAT basically bakes the compression directly into the training process. Instead of standard post training quantization degrading the model's reasoning capabilities, QAT trains the model with compression in mind. Unsloth is reporting near original performance at 4-bit with ~72% lower memory footprint. Details in the comments. Naturally, had to instantly pull the new GGUFs to see what a single RTX 4090 card (24 GB VRAM, Cuda 12.8, ubuntu 22) could do. i fired up llama.cpp engine again Look at these numbers: 1. Unsloth Gemma 4 26B-A4B IT (QAT Q4_K_XL) flags: ./build/bin/llama-cli -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 250000 -fa on -v VRAM Used: 19.5 GB context: 250,000 tokens decode throughput: 193 tps 2. Unsloth Gemma 4 31B IT (QAT Q4_K_XL) flags: Command: ./build/bin/llama-cli -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 60000 -fa on -v - VRAM Used: 23 GB (Tight, but zero system RAM spillover) - context: 60,000 tokens - decode throughput: 47 tps We are essentially watching hardware bottlenecks evaporate in real time. An update literally drops before you can finish benchmarking the previous one. What a time to be running local hardware. If you have a single rtx 3090, rtx 4090, these are the latest gemma models to try this week.show more

Ever wished your AI coding agent had perfect memory?... 🧠 Memory Bank is a game-changing approach that gives AI persistent memory across coding sessions. Here's how it works and why you should be using it: 🧵/show more

Qwen3.8-Flash-Next now reaches ~43 tok/s after a 122,902-token prompt... on ONE DGX Spark. ⚡🚀 MTP k=2 won my draft-depth sweep, with +42.5% mean decode over no draft. The PLE table stays fully on-device. I promised the deeper MTP tests. Here are the results, and now you can explore them in an interactive benchmark page too. 𝗧𝗪𝗢 𝗗𝗥𝗔𝗙𝗧 𝗧𝗢𝗞𝗘𝗡𝗦 𝗪𝗢𝗡 Mean single-request decode with 32K context configured: MTP k=2: 39.21 tok/s MTP k=3: 36.42 tok/s MTP k=1: 35.18 tok/s No draft: 27.51 tok/s k=2 also produced the fastest individual sweep run: 41.34 tok/s. Four runs each for no draft, k=1 and k=2. Seven for k=3. Decode excludes time to first token. Here, k means speculative draft depth, not quantization bits. k=3 produced more tokens per step, but the extra drafting work did not pay off in throughput. k=2 is my current pick for this setup. 𝗧𝗛𝗘 𝟭𝟮𝟯𝗞-𝗧𝗢𝗞𝗘𝗡 𝗣𝗥𝗢𝗠𝗣𝗧 𝗧𝗘𝗦𝗧 I then ran a separate long-prompt comparison: Actual input: 122,902 tokens Configured context: 262,144 Requested output: 128 tokens One request at a time MTP k=2: ~43 tok/s No draft: 26.2 tok/s Time to first token: 110.6 seconds with MTP 107.0 seconds without it The win here is generation speed, not faster prefill. To keep the scope clear: 256K was the configured limit. This was a real ~123K input, not a completely filled 256K window or a full k sweep at that depth. 𝗣𝗟𝗘 𝗦𝗧𝗔𝗬𝗦 𝗢𝗡 𝗧𝗛𝗘 𝗦𝗣𝗔𝗥𝗞 Whole model on-device: 78.57 GiB Packed 5-bit PLE table: 30.4 GiB, included in that total No NVMe PLE offload in this build. This is still turboderp’s 3.05bpw_h5_ng5 EXL3 pack, served through my vllm-exl3 integration. My work here is the serving integration and testing. These are preliminary performance measurements, not a quality evaluation or a claim of bit-exact full-output parity. 𝗘𝗫𝗣𝗟𝗢𝗥𝗘 𝗧𝗛𝗘 𝗥𝗘𝗦𝗨𝗟𝗧𝗦 The benchmark page has the individual sweep values, long-prompt comparison, and measurement scope. You can play the animation, export the charts, or download the HTML and data to render them yourself. No Spark needed to view the results. Credit to turboderp / ExLlamaV3 for the pack and kernels, vLLM for the serving engine, and Qwen Qwen Developers for the model. Recipe + reproduction: Interactive benchmark:show more

8GB RTX 4060 Laptop. ~16GB Ornith IQ3_S GGUF. 44.5... tok/s streaming locally. FreeToken didn’t have the qwen35moe GGUF path needed for this model, so I implemented it and submitted PR #131 upstream. 👀 I also exposed the K/I GGUF quant types its CUDA kernels already supported through the Python loader. This is the result. Setup: Ornith-1.5-35B-A3B 35B MoE / ~3B active IQ3_S GGUF · ~16GB RTX 4060 Laptop · 8GB VRAM FreeToken · FlashInfer MoE backend: offload On the attached full-prompt run, I measured ~44.5 tok/s streaming on screen. In the same runs, FreeToken’s server decode counter reported 46.7 → 50.1 tok/s. VRAM stayed at 6,879 / 8,188 MiB, with 20GB host RAM and 98% GPU utilization during generation. Load time: 65s. My llama.cpp CPU reference on the same model was 11.1 tok/s. The interesting part isn’t just “35B on an 8GB GPU.” The code path itself was missing. FreeToken’s CUDA side already supported more GGUF quant types than its Python loader exposed, including the K/I quants needed for this IQ3_S model. And qwen35moe models like Ornith needed their own GGUF architecture/loading path. I implemented both pieces in my fork and opened them upstream in PR #131. Once loaded, FreeToken can use GPU + CPU + host RAM as an elastic MoE inference system instead of requiring the entire checkpoint to live in VRAM. That fits Ornith well because only a subset of its experts are active per token. Important caveat: PR #131 is still open. This is not merged upstream yet. I’m not calling this official Ornith support. I’m showing a working implementation, the code is public, and the attached terminal video is the actual run. PR #131: My fork: Next I’m going to test where this path holds up across more quants, longer context and real agent workloads.show more

The "I don't have enough VRAM" excuse just died.... I’m running Meta’s new 30B Muse Glimmer Q6_K_XL with a massive 130k context window on just 26GB VRAM FREE compute on Kaggle. Kaggle provides you free 2x Nvidia T4 GPUs. 30 hours usage each week! Yesterday, I showed you the violent throughput of Muse Glimmer on a single RTX 4090. Today, we are securing a Dual NVIDIA T4 GPU cluster with 32GB of total VRAM for exactly $0 and dropping the massive 24.5GB Q6_K_XL GGUF onto it. Here is the exact Kaggle workflow and benchmarking breakdown: # 1. The Storage Bypass & Setup I built a clean cell by cell script in the file. We dynamically fetch the CUDA accelerated llama.cpp binaries and use wget to stream the model directly into Kaggle's /kaggle/tmp scratch storage, which cleanly bypasses their 19.5GB output directory limit. # 2. The Multi GPU Performance With the -ngl 99 flag offloading all model layers across both T4 GPUs (32GB VRAM combined), we pushed a massive 131,072 token context window (-c 131072). The benchmark numbers: Prefill: 265.9 t/s Decode: 9.0 t/s VRAM Total: 26.5 GB # 3. The Architecture Insight The Q6_K_XL model itself is 24.5 GB. Because of Muse Glimmer's aggressive 16:1 GQA, the unquantized KV cache for a massive 130k context window only takes up 2 GB of memory. No heavily degraded Q4 KV quantization required. It just works. No compiling from source. No credit card. No OOM crashes. Zero excuses. If you’re running a single RTX 3090, 4090, or 5090, you need to experience this hyper efficient KV cache right now before the upcoming Qwen 3.8 27B drop completely steals your VRAM tomorrow. pick the Q4 or Q5 quants for 24 GB VRAM rigs. I'm dropping the Unsloth huggingface GGUF links and the free Kaggle notebook link in the replies. spin up your own instance, and show me your multi GPU benchmarks.show more

We are in an insane run of open-weight drops.... Every modality, open source is winning. This is what an open source AI summer ☀️ looks like: 🧠 LLMs & Reasoning → DeepSeek-V4-Flash-0731 (my king 👑): 304B MoE refresh, Terminal-Bench 2.1 jumps 61.8→82.7 over the preview, DeepSWE 7.3→54.4. Closes in on Opus-4.8 on Agents' Last Exam (25.2 vs 25.7). MIT. → Muse-Glimmer-30B, from Meta (they are back!!): their first open agentic model. ~29.6B dense + perception encoder, 131k+ context, built to run fully local, no cloud. Apache 2.0. → Liquid AI LFM2.5-2.6B: 2.69B params, 131k context, 220 tok/s on an M5 Max in under 2.5GB RAM. Competitive with models 4x larger on agentic tasks. → inclusionAI Ling-3.0-flash: 124B total, only 5.1B active, ~12% the size of their old 1T flagship Ring-2.6, matches it on key benchmarks. MIT. → inclusionAI Ling-3.0-tiny: 7.9B total, 1.3B active, 86-90 tok/s on an M4 Pro MacBook at ~8GB peak memory. MIT. → NVIDIA Nemotron-3.5-Lightning-30B-A3B: hybrid Mamba-2+MoE+Attention, up to 1M context, runs on a single H100 or DGX Spark, SWE-bench Verified 52.8. → deepgrove maple-preview: 20B-A1B ternary-weight reasoner, 218 tok/s on a Mac mini M4, 5.3GB checkpoint. MIT. → BigBang-v1 (endless-frontier): fine-tuned from Qwen3.6-35B-A3B via a self-evolving generator/critic synthetic-data loop. Lands aggregate performance between DeepSeek V4 Flash (284B) and V4 Pro (1.6T), at 35B. Apache 2.0. 🎬 Video → MiniMax-H3: 33B dense omni model, native stereo audio, up to 2K/15s. 3.6k+ likes already. → Minimax-H3-Turbo (lightx2v): Apache-2.0 turbo distillation of H3 for fast inference. → Lightricks LTX-2.5: image-to-video update, custom Gemma-4-12B text encoder, a markedly stronger distilled model. 🔊 Voice → NVIDIA NemotronLabs VoiceChat-11B: full-duplex speech-to-speech, ~450ms turn-taking, #2 on open VoiceBench, and the first open full-duplex model with live tool-calling mid-conversation. 🛡️ Safety → Mistral Shieldstral-1.0-3B: 3B multimodal guardrail that takes your safety policy as plain text instead of fixed categories. Beats LlamaGuard-4-12B and ShieldGemma-9B on HarmBench (99.4) and ToxicChat (84.1) at a fraction of the size. Apache 2.0.show more