---
title: Local AI
url: "https://goremy.ai/blog/topic/local-ai"
type: topic
section: AI News & Models
updated: "2026-11-12T12:40:15.148Z"
posts: 150
---

# Local AI

Run AI models on your own laptop, GPU or server: setup guides, hardware, benchmarks and when local beats the cloud.

Answers: How much VRAM do I need to run open-weight models locally?

## Posts (150)

- [Can You Run Ripple 65B Locally? Hardware and Comparison](https://goremy.ai/blog/articles/ripple-65b-preview-run-locally) (2026-11-12): Ripple 65B Preview is a dense 65B open-weight model with 1M context. See memory needs, DGX Spark fit, API pricing and how it compares to peers.
- [How to Run Underdog Woof TTS on a CPU (and Is It Free to Use?)](https://goremy.ai/blog/articles/how-to-run-underdog-woof-tts-on-cpu) (2026-11-12): Underdog Woof is a 40.6 MB English text-to-speech model that runs on a CPU with NumPy. Setup steps, voices, test results and the non-commercial license catch.
- [Jensen Huang and Megaspeed: The Nvidia Chip Probe Explained](https://goremy.ai/blog/articles/nvidia-jensen-huang-china-chips-megaspeed-investigation) (2026-11-12): Nvidia's board is probing a Jensen Huang relationship tied to Megaspeed while 600,000 H200s wait in Hong Kong. The timeline, export rules and stakes.
- [How Good Is Qwen4 Max? Hands-On Tests vs Qwen 3.8](https://goremy.ai/blog/articles/qwen4-max-hands-on-tests) (2026-11-12): Qwen4 Max preview tested on Blender animation, an 82-language naming task, satellite map reasoning and seat belt laws, plus where it struggled.
- [Underdog Woof TTS Review: How Accurate Is It, Really?](https://goremy.ai/blog/articles/underdog-woof-tts-test-results-accuracy) (2026-11-11): Underdog Woof claims a 1.42% word error rate. Hands-on tests of Juno and Mira on phone numbers, prices, titles and dates show where it slips.
- [Nemotron Bolt vs Super: How Much Quality Does Bolt Lose?](https://goremy.ai/blog/articles/nemotron-bolt-vs-super-benchmarks-agent-tests) (2026-11-11): Nemotron 3.5 Bolt is a 2-bit version of the 120B Super. Compare benchmark drops (PinchBench 90 to 88, GPQA 89 to 86), file size, speed and what to test.
- [How Much Does Fable 5.2 Cost? Real Token Bills by Project](https://goremy.ai/blog/articles/fable-5-2-ultra-code-token-costs) (2026-11-11): Real token counts and API-equivalent costs for Fable 5.2 projects, from a $30 Blender house to a $15,000 game, plus Ultra Code and the usage reset.
- [How to Run TwIL LM3 Pro Locally With llama.cpp](https://goremy.ai/blog/articles/twil-lm3-pro-run-locally-llama-cpp) (2026-11-11): Run the 3.66B TwIL LM3 Pro logic model locally with llama.cpp: VRAM needs, quantized builds, a Hermes agent setup, and the non-commercial license.
- [How Good Is TwIL LM3 Pro at Logic and Contradictions?](https://goremy.ai/blog/articles/twil-lm3-pro-logic-test-results) (2026-11-11): Hands-on results for TwIL LM3 Pro: 3/3 on spec contradictions, 8/10 on a return policy, 5/8 on invented-rule puzzles, and where larger models win.
- [Can You Run Nvidia Nemotron 3.5 Bolt 74B on 24GB?](https://goremy.ai/blog/articles/nvidia-nemotron-3-5-bolt-74b-2-bit-local) (2026-11-10): Nvidia's Nemotron 3.5 Bolt 74B A9B stores experts at 2 bits to fit in 24GB. See the memory math, speed on RTX Spark, and what you give up.
- [Tencent HY4: Pricing, Vision, Coding Tests and Open Weights](https://goremy.ai/blog/articles/tencent-hy4-tested-pricing-open-weights) (2026-11-10): Tencent HY4 hands-on: coding tests, new vision support, $2.50 output pricing versus DeepSeek V4.1 Flash, and whether open weights are coming.
- [How to Run Arete-27B Locally: VRAM, Quantization, Context](https://goremy.ai/blog/articles/arete-27b-run-locally-requirements) (2026-11-10): Run Nous Research's Arete-27B locally: VRAM estimates by quantization, context limits from its 64k training, and runtime settings to start with.
- [How to Install DeepTutor with Ollama as a Local AI Tutor](https://goremy.ai/blog/articles/deeptutor-ollama-local-ai-tutor-setup) (2026-11-10): Learn to install HKUDS DeepTutor in Docker, connect Ollama chat and embedding models, and test its quiz, solve, research and visualize modes.
- [How to Run Arete-27B Locally: Hardware and Quantization](https://goremy.ai/blog/articles/how-to-run-arete-27b-locally) (2026-11-10): Hardware needed to run Nous Research's Arete-27B locally: VRAM by quantization level, the 64k training context, text-only limits, and runtime choices.
- [How to Run NVIDIA RE-USE Speech Enhancement on a CPU](https://goremy.ai/blog/articles/nvidia-nemotron-re-use-speech-enhancement-cpu) (2026-11-10): Run NVIDIA's 9.6M-parameter RE-USE model on CPU to remove noise, echo and clipping from speech. Setup commands, test results, music limits, architecture.
- [How Well Does AREX-2 Score? MLE-Bench Lite and Skills Ablation](https://goremy.ai/blog/articles/arex-2-benchmarks-mle-bench-skills-ablation) (2026-11-10): AREX-2 scores 81.8 on MLE-Bench Lite and 70.7 on FrontierCS. See the 5-hour curve and how much of the gain comes from skills versus training.
- [How Does MiniCPM-OCR Score Against Other Open OCR Models?](https://goremy.ai/blog/articles/minicpm-ocr-benchmarks-vs-other-open-ocr-models) (2026-11-10): MiniCPM-OCR leads four of six benchmarks against 15 open OCR models and runs fast, but charts, reading order and footers still trip it up.
- [How to Run MiniCPM-OCR 1.0 Locally: VRAM and Test Results](https://goremy.ai/blog/articles/minicpm-ocr-1-0-run-locally-test-results) (2026-11-10): How to run OpenBMB's MiniCPM-OCR 1.0 locally with vLLM, what VRAM it needs, and how it handled formulas, charts, newsprint, invoices and Chinese contracts.
- [How Do You Control an Olares One With a Coding Agent?](https://goremy.ai/blog/articles/olares-cli-control-local-ai-server-with-codex) (2026-11-10): How a coding agent like Codex or Claude Code can drive an Olares One: what the setup involves, example prompts for STL files and ComfyUI, and what to verify.
- [Can a Local Qwen 27B Design Hardware? ESP32 GPS Build Test](https://goremy.ai/blog/articles/vibe-coding-hardware-local-qwen-olares-one) (2026-11-10): A 4-bit Qwen 3.8 27B on a 24GB Olares One drove OpenCode to design firmware and OpenSCAD enclosures for an ESP32 GPS speedometer. Here is what held up.
- [Can You Run Local AI on a Weak Linux Machine With Low RAM?](https://goremy.ai/blog/articles/local-ai-weak-linux-machine-limited-memory) (2026-11-09): Run local AI on a weak Linux machine with 4GB RAM and no GPU: the RAM budget, model sizes that fit, Ollama setup, and realistic CPU-only speeds.
- [How to Run StarCoder2 15B Locally: Hardware and Commands](https://goremy.ai/blog/articles/run-starcoder2-15b-locally) (2026-11-09): Run StarCoder2 15B locally with Ollama or llama.cpp. Install commands, approximate quantized sizes, and the VRAM or unified memory each setup needs.
- [How Do You Run DSpine With Qwen3.8-27B on SGLang?](https://goremy.ai/blog/articles/run-dspine-qwen3-8-27b-sglang-guide) (2026-11-09): Set up Tencent's DSpine draft model with Qwen3.8-27B on SGLang: install steps, serve flags, block size 16, H100 memory use and the A100 problem.
- [DSpine vs DFlash2: How Much Faster Is It? Benchmarks](https://goremy.ai/blog/articles/dspine-vs-dflash2-benchmarks-acceptance-length-throughput) (2026-11-09): DSpine vs DFlash2 on Qwen3.8-27B: acceptance length, agent throughput, and a hands-on H100 test that hit about 156 tokens/s versus 33.
- [How to Run the Edited Qwen3.5 0.8B Locally With vLLM](https://goremy.ai/blog/articles/run-weight-edited-qwen3-5-0-8b-vllm) (2026-11-07): Serve Hugging Face's rebuilt weight-edited Qwen3.5 0.8B beside the original in vLLM and compare looping on a 2^100 mod 7 prompt and MATH-500 problems.
- [Northka Computer Use Model: Which Size and How Good?](https://goremy.ai/blog/articles/northka-open-computer-use-model-osworld-scores) (2026-11-07): Northka is an open-weight computer-use model family (397B Pro, 35B Mini, 9B Nano) scoring 78.9 to 88 on OSWorld Verified. Sizes, scores and harness explained.
- [How Do You Fix a Small Reasoning Model Stuck in Loops?](https://goremy.ai/blog/articles/fix-small-reasoning-model-looping-truncated-answers) (2026-11-06): Small reasoning models often loop until the token limit cuts them off. Count truncated answers first, then adjust mode, temperature and output limit.
- [Can You Run Qwen 3.8 27B Locally With Ollama? VRAM Needs](https://goremy.ai/blog/articles/run-qwen-3-8-27b-ollama-read-model-weights) (2026-11-06): Run Qwen 3.8 27B locally with Ollama and Simon Willison's LLM tool. Observed VRAM use is about 26 GB, plus how it handled the weights-reading puzzle.
- [Did GPT-6 Astra Really Make Qwen 6x Better at Math?](https://goremy.ai/blog/articles/gpt-6-astra-weight-space-programming-qwen-math) (2026-11-06): An MIT-linked paper says GPT-6 Astra edited 0.016% of Qwen3.5-0.8B's weights for a sixfold math gain. Here is what changed and what to doubt.
- [How to Run Ripple-65B Locally: VRAM and Setup Requirements](https://goremy.ai/blog/articles/how-to-run-ripple-65b-locally-vram) (2026-11-06): What it takes to self-host Ripple-65B: about 75 GB VRAM quantized on a 96 GB RTX Pro 6000 Blackwell, inference code changes, a Hermes agent, and DGX Spark fit.
- [Ripple-65B Tested: How Good Is This Dense Open Model?](https://goremy.ai/blog/articles/ripple-65b-dense-model-tested-agentic-coding) (2026-11-05): Ripple-65B is a 65B dense Apache 2.0 model aimed at agentic coding. See its benchmark wins and gaps, plus a bug-fix, simulation and SQL test.
- [How Good Is Tencent Hy4 at Long Agentic Tasks? Hands-On Tests](https://goremy.ai/blog/articles/tencent-hy4-hands-on-test-blender-finance) (2026-11-05): Tencent's Hy4 was tested on a sketch-to-Blender scene, a multi-agent IPO report with a planted error, and 82-language output. Results and caveats.
- [How to Run Qwen3.5-4B GLB 3D Coder Locally and Use It in Godot](https://goremy.ai/blog/articles/qwen3-5-4b-glb-3d-coder-run-locally-godot) (2026-11-04): Run CCORE's Qwen3.5-4B GLB 3D Coder locally from a GGUF, set the sampling and system prompt, then export Three.js scenes as GLB files for Godot.
- [How Do You Run NVIDIA Dory Locally? Loops, KV Cache, Formats](https://goremy.ai/blog/articles/how-to-run-nvidia-dory-locally-vllm) (2026-11-03): Run NVIDIA's 3.97B Dory looped model locally: pick a loop count in vLLM, choose per-loop or last-loop KV cache, and pick BF16, MXFP8, NVFP4 or GGUF.
- [Frontier vs Open-Weight Model: How to Decide With Step-Down Tests](https://goremy.ai/blog/articles/how-to-choose-open-weight-vs-frontier-model-step-down-test) (2026-11-03): How to decide if you can replace a frontier model with a self-hosted open-weight one: build a swap layer, step down generation by generation, then fine-tune.
- [How to Run MOSS-Image 5B Locally in ComfyUI (VRAM, Setup)](https://goremy.ai/blog/articles/moss-image-5b-comfyui-local-install) (2026-11-03): Install the 5B MOSS-Image text-to-image model in ComfyUI: where the files go, the node setting that matters, about 28 GB VRAM use and layout mode.
- [How to Install Mesh LLM and Pool GPUs Across Machines](https://goremy.ai/blog/articles/mesh-llm-pool-gpus-openai-compatible-api) (2026-11-03): Install Mesh LLM with Docker, start a node with a Gemma model, edit and validate config, and use one OpenAI-compatible API across pooled GPUs.
- [How to Run SenseNova-Loop-9B-Preview: Setup and Serving](https://goremy.ai/blog/articles/how-to-run-sensenova-loop-9b-preview) (2026-11-03): Run SenseNova-Loop-9B-Preview with Transformers 5.18, LightLLM or an SGLang PR build: environment, BF16 loading, thinking mode, MTP and fine-tuning commands.
- [MOSS-Image Review: Layout Control and Text Rendering Tested](https://goremy.ai/blog/articles/moss-image-layout-control-text-rendering-test) (2026-11-03): Hands-on MOSS-Image results: bounding-box layout control, English, Chinese and Urdu text, plus portraits and landscapes from many regions.
- [LFM3-1.7B-Base: Specs, Memory Needs and Fine-Tuning Guide](https://goremy.ai/blog/articles/lfm3-1-7b-base-specs-local-fine-tuning) (2026-11-02): Liquid AI's LFM3-1.7B-Base is a 1.73B model with 256K context that takes about 1 GB at 4-bit. Specs, inference code, thinking toggle and fine-tuning with TRL.
- [How to Run DeepSeek and Qwen on One GPU With Colibri 2.1](https://goremy.ai/blog/articles/colibri-2-1-run-gguf-deepseek-qwen-one-gpu) (2026-11-02): Colibri 2.1 runs large MoE models on one GPU by spreading experts across VRAM, RAM and disk. See the install flow, GGUF import, expert map and profiling tab.
- [Why Did Open Models Fall 12 Points Behind in Index v5?](https://goremy.ai/blog/articles/artificial-analysis-index-v5-open-models-gap) (2026-11-02): Artificial Analysis Intelligence Index v5 uses 60% private tests and harder agentic coding. The open-model gap grew from 3 to 12 points. Here is what it means.
- [Shared On-Prem AI Box vs Cloud APIs: When Does It Win?](https://goremy.ai/blog/articles/on-prem-ai-workstation-vs-cloud-api-team-fury) (2026-11-01): When does one shared on-prem GB300 workstation beat cloud APIs or a Spark per person? Privacy, token costs, model policy control, and scaling.
- [How Accurate Is Phonon 2? 5.21% WER in a 164MB ASR Model](https://goremy.ai/blog/articles/phonon-2-word-error-rate-open-asr-benchmark) (2026-11-01): Phonon 2 averages 5.21% WER on Open ASR leaderboard English sets in 164MB, 15x smaller than Parakeet. See where it holds up and where it fails.
- [How to Run Phonon 2 Locally on CPU for Speech Recognition](https://goremy.ai/blog/articles/phonon-2-speech-recognition-run-locally-cpu) (2026-11-01): Run Fermion Research's 164 MB Phonon 2 ASR model on CPU only. Install steps, CPU timings, the OpenAI-compatible server, accuracy numbers and license notes.
- [How to Run Surya OCR 2.5 Locally on CPU or GPU](https://goremy.ai/blog/articles/run-surya-ocr-2-5-locally) (2026-11-01): Install Surya OCR 2.5 with pip, launch the Streamlit GUI, and run it on CPU or about 1.8 GB of VRAM. Covers backends, output formats, speed, and limits.
- [How Accurate Is Surya OCR 2.5? Tested on Real Documents](https://goremy.ai/blog/articles/surya-ocr-2-5-test-results-languages) (2026-10-31): Surya OCR 2.5 tested on invoices, handwriting, an 1883 newspaper, LaTeX, Hindi, Arabic, Urdu and Yoruba. See where it works and where it fails.
- [HP ZGX Fury Review: Can One GB300 Box Serve a Whole Team?](https://goremy.ai/blog/articles/hp-zgx-fury-gb300-local-ai-workstation-review) (2026-10-31): HP's ZGX Fury puts Nvidia's GB300 under a desk with 748GB of memory and 7.1 TB/s HBM. Here are the specs, ZRT software and measured multi-user throughput.
- [How to Distill a Small LLM: Tokenizers, Pass@4, and Pitfalls](https://goremy.ai/blog/articles/how-to-distill-small-model-on-policy-distillation-guide) (2026-10-31): A practical guide to distilling a large LLM into a small one: the four-tries headroom test, tokenizer matching, vLLM memory limits, and when to stop training.
- [What GPU Does Sage A1.5 Need? Local VRAM and Agent Tests](https://goremy.ai/blog/articles/sage-a1-5-35b-agentic-model-local-test) (2026-10-30): Sage A1.5 is a 35B MoE agentic model from Intern. See the reported 38 GB VRAM with llama.cpp, how it was trained, and how it handled sub-agent tasks.
- [Is Clover-35B-A3B-Preview Any Good? Diffusion LLM Tested](https://goremy.ai/blog/articles/clover-35b-a3b-preview-diffusion-llm-review) (2026-10-30): Hands-on test of DiffuSpace's Clover-35B-A3B-Preview diffusion LLM: bug fix, coding, multilingual and logic puzzle results versus its benchmark claims.
- [How to Run Clover-35B-A3B-Preview Locally: VRAM and Settings](https://goremy.ai/blog/articles/run-clover-35b-a3b-preview-locally) (2026-10-30): Run DiffuSpace's Clover-35B-A3B diffusion model locally: about 69 GB of BF16 weights, Transformers 5.19.0 setup, block diffusion defaults, controls and SGLang.
- [Does LFM3-1.7B Match 9B Models on Agent Benchmarks?](https://goremy.ai/blog/articles/lfm3-1-7b-benchmarks-vs-qwen-gemma-minicpm) (2026-10-29): LFM3-1.7B is pitched as matching 9B models on agent benchmarks. Here is what the card claims, what hands-on tests showed, and where it fell short.
- [How to Fine-Tune LFM3-1.7B With TRL on a Single GPU](https://goremy.ai/blog/articles/fine-tune-lfm3-1-7b-trl-sft-dpo-grpo) (2026-10-29): Learn how to fine-tune Liquid AI's LFM3-1.7B with TRL using SFT, DPO, GRPO and continued pre-training, and which recipe fits JSON, RAG, tools or new languages.
- [How to Run Baidu Unlimited OCR 2 Locally: VRAM and Results](https://goremy.ai/blog/articles/baidu-unlimited-ocr-2-run-locally) (2026-10-29): How to run Baidu's Unlimited OCR 2 on your own GPU: about 8.3 GB VRAM at full precision, Transformers and SGLang setup, and test results on real documents.
- [How to Run LFM3-1.7B Locally: Hardware, Formats, Settings](https://goremy.ai/blog/articles/lfm3-1-7b-liquid-ai-local-agent-model) (2026-10-28): Run Liquid AI's LFM3-1.7B on a phone or laptop: GGUF, MLX, ONNX, FP8 and NVFP4 checkpoints, sampling settings, tool calling, and where it falls short.
- [How Does Unlimited OCR 2 Score on 100-Page Documents?](https://goremy.ai/blog/articles/unlimited-ocr-2-long-document-benchmark-results) (2026-10-28): Unlimited OCR 2 scores about 77 on a 100-page document test while most rivals fall under 30. How SR-SWA attention works, plus where the model still fails.
- [How to Run DeepSeek V4.1 Flash GGUF on 12GB VRAM Locally](https://goremy.ai/blog/articles/deepseek-v4-1-flash-gguf-run-under-12gb-vram) (2026-10-28): Run 2-bit GSQ-RCO GGUF builds of DeepSeek V4.1 Flash on a 12GB GPU. Covers file sizes, Strata vs llama.cpp support, MoE design and benchmark recovery.
- [What Are GSQ and RCO Quantization? 2-Bit GGUF Explained](https://goremy.ai/blog/articles/gsq-rco-quantization-explained) (2026-10-28): GSQ learns how to round each weight and RCO decides how many bits each layer gets. See how the two shrink a huge MoE model to about 2 bits in a standard GGUF.
- [How Much VRAM Does Nanbeige4.3 33B Need Running on vLLM?](https://goremy.ai/blog/articles/run-nanbeige4-3-33b-vllm-vram-requirements) (2026-10-28): Nanbeige4.3 33B used just under 88 GB of VRAM on vLLM at long context, about 62 GB at full precision. Here is how to size hardware and cut memory.
- [How Do You Run Moss Image Locally in ComfyUI or SGLang?](https://goremy.ai/blog/articles/run-moss-image-locally-comfyui-sglang-loras) (2026-10-28): Run the 5B Moss Image model in ComfyUI, SGLang or vLLM-Omni. Covers SFT vs base checkpoints, 50-step defaults, 8-step LoRAs and quantized builds.
- [Is Nanbeige4.3 33B Claude-Level? Benchmarks vs Real Tests](https://goremy.ai/blog/articles/nanbeige4-3-33b-test-results-claude-level-claim) (2026-10-27): Nanbeige4.3 33B tops Qwen 3.8 27B and Gemma 4 on most charts, but looped endlessly in a bug-fix and an HTML test. Here is how the claims hold up.
- [How Good Is Moss Image 5B, the Apache 2.0 Image Model?](https://goremy.ai/blog/articles/moss-image-5b-open-text-to-image-review) (2026-10-27): Moss Image is a 5B open-weight text-to-image model under Apache 2.0. See its text rendering, layout control, benchmark claims, and where it falls short.
- [Kimi K3.1 Pro vs Opus 5.5: Has Open Source Caught Up?](https://goremy.ai/blog/articles/kimi-k3-1-pro-vs-opus-5-5-open-source-caught-up) (2026-10-27): Kimi K3.1 Pro ranks fourth on Artificial Analysis and sits under a point behind Opus 5.5 on a coding benchmark. Here is what that means for builders.
- [How to Run Nemotron-3.5 Bolt E4B Locally With llama.cpp](https://goremy.ai/blog/articles/run-nemotron-3-5-bolt-e4b-locally) (2026-10-27): Run NVIDIA Nemotron-3.5 Bolt E4B locally with llama.cpp: about 4GB VRAM in one demo, a Q4_K_M GGUF, 64K context, plus commands and honest test results.
- [How Good Is Nemotron-3.5 Bolt E4B? Tested Against Gemma 4](https://goremy.ai/blog/articles/nemotron-3-5-bolt-e4b-hands-on-test-results) (2026-10-27): Hands-on results for NVIDIA's Nemotron-3.5 Bolt E4B: a bug-hunt agent task that failed, a broken HTML scene, and passes on a riddle, physics math and rostering.
- [How Does Nanbeige4.3-33B Score vs Qwen3.8-27B and Gemma4-31B?](https://goremy.ai/blog/articles/nanbeige-4-3-33b-benchmarks-vs-qwen-gemma) (2026-10-27): Nanbeige4.3-33B leads 17 of 20 benchmarks against Qwen3.8-27B, Gemma4-31B and Muse-Glimmer-30B. See the full table and where it loses.
- [Kimi K3.1 Pro vs Opus 5.5: Which Is Better for Agentic Tasks?](https://goremy.ai/blog/articles/kimi-k3-1-pro-vs-opus-5-5-claude-code) (2026-10-27): A 13-task test of Kimi K3.1 Pro against Claude Opus 5.5 in the same Claude Code harness: where Kimi matches Opus, where it cuts corners, and what to check.
- [How to Run FLUX 3 Klein Locally in ComfyUI: VRAM and License](https://goremy.ai/blog/articles/flux-3-klein-local-install-comfyui-vram-license) (2026-10-26): Run FLUX 3 Klein locally in ComfyUI: 5.6B transformer, 16-step distilled model, about 26.4GB VRAM at full precision, and the under-$5M commercial license.
- [How to Run Nanbeige4.3-33B Locally With vLLM or SGLang](https://goremy.ai/blog/articles/how-to-run-nanbeige-4-3-33b-vllm-sglang) (2026-10-26): Run Nanbeige4.3-33B with Transformers, vLLM or SGLang. Covers the 512K context, thinking flags, tool-call format, and recommended sampling settings.
- [How to Enable Faster Prefill on RTX 3090 and 4090 for vLLM](https://goremy.ai/blog/articles/rtx-3090-gpu-unlocker-prefill-speedup-vllm) (2026-10-26): A free GPU unlocker patch speeds up prefill on RTX 3090, 4090 and other Ampere/Ada cards in vLLM. Steps for Proxmox/Linux plus before and after numbers.
- [Kimi K3.1 Pro vs Opus 5.5 vs GPT 6.1 Soul: Which Is Better?](https://goremy.ai/blog/articles/kimi-k3-1-pro-vs-opus-5-5-gpt-6-1-soul) (2026-10-25): Kimi K3.1 Pro tested against Opus 5.5 and GPT 6.1 Soul: pricing, open-weight caveats, and hands-on demos from a Rubik's cube to a Fall Guys clone.
- [Is Local AI Worth It? Cost vs ChatGPT and Claude Plans](https://goremy.ai/blog/articles/is-local-ai-worth-it-cost-vs-chatgpt-claude) (2026-10-25): Break-even math on local AI: a $4,000 machine over three years versus ChatGPT or Claude plans, plus what data sovereignty and access are worth.
- [Kimi K3.1 Pro Benchmark: Does It Match Claude Opus 5.5?](https://goremy.ai/blog/articles/kimi-k3-1-pro-kingbench-results) (2026-10-25): Kimi K3.1 Pro claims near-Opus results, but a six-task KingBench run scored it 5.37/10. See task scores, cost, and how it compares to rivals.
- [Can You Run Perplexity Computer Locally on a Mac? (36GB)](https://goremy.ai/blog/articles/run-perplexity-computer-locally-on-mac) (2026-10-24): Perplexity Computer is reported to run on a Mac with Qwen 3.6 instead of cloud machines. What the 36GB unified memory figure means and what to check first.
- [Local LLM Hardware: What Tokens Per Second Can You Expect?](https://goremy.ai/blog/articles/local-ai-gpu-mac-studio-tokens-per-second) (2026-10-24): Local LLM hardware tiers compared: RTX 4090, RTX 5090, RTX Pro 6000 Blackwell and Mac Studio, with prices, tokens per second, and multi-agent limits.
- [Flint vs Qwen 3.8 27B: Which Local Coding Model Wins?](https://goremy.ai/blog/articles/flint-vs-qwen-3-8-27b-coding-benchmarks) (2026-10-24): Flint is roughly 10x faster than dense Qwen 3.8 27B but scores 55 vs 73 on Terminal Bench. Here is where each local coding model fits.
- [How Do You Run Carbon-X, the 14B DNA Language Model, Locally?](https://goremy.ai/blog/articles/carbon-x-dna-language-model-run-locally) (2026-10-24): Carbon-X is a 14.8B open DNA and text model that fits on one 80 GB GPU. Set up Transformers v5 or vLLM, handle DNA tokens, and call it from an agent.
- [How Good Is Ling-3.0 Flash Santé at Medical Reasoning?](https://goremy.ai/blog/articles/ling-3-0-flash-sante-healthcare-model-tested) (2026-10-24): InclusionAI's MIT-licensed Ling-3.0 Flash Santé is a 124B MoE healthcare model. See clinical test results, benchmark standing, and its lack of vision.
- [How Accurate Is Carbon-X on DNA Tasks vs GPT and Opus?](https://goremy.ai/blog/articles/carbon-x-benchmarks-vs-gpt-opus-dna-variant-effects) (2026-10-23): Carbon-X, a 14.8B DNA model, averages 66.7% on five benchmarks versus 67.5% for GPT-6.1 Sol at far lower cost. Results, weak spots, and limits.
- [How Good Is Flint 40B at Coding? Benchmarks vs Qwen 3.8](https://goremy.ai/blog/articles/flint-40b-benchmarks-terminal-bench-swe-bench) (2026-10-23): Flint 40B scores 54.9% on Terminal-Bench 2.1 and 74.2% on SWE-bench Verified. See M4 Pro speed, NVFP4 vs BF16, and where Qwen 3.8 still wins.
- [How Does Sage-A1.5 Compare to Kimi-K3, GLM-5.3 and GPT-5.6-sol?](https://goremy.ai/blog/articles/sage-a1-5-benchmarks-vs-kimi-k3-glm-5-3) (2026-10-23): Sage-A1.5, a 35B MoE agentic model, hits 95.1 on GAIA and 61.2 on SWE-bench Pro. Here is where it beats or trails Kimi-K3, GLM-5.3 and GPT-5.6-sol.
- [How to Run Flint 40B Locally: Ollama, GGUF, SGLang and VRAM Needs](https://goremy.ai/blog/articles/how-to-run-flint-40b-locally-ollama) (2026-10-23): Install Flint 40B locally with Ollama, llama.cpp or SGLang. Covers VRAM needs (about 32GB for the quant), GGUF Q4, NVFP4, MLX and BF16 on H200 or B300.
- [How Do You Serve Sage-A1.5 With SGLang or vLLM at 512K Context?](https://goremy.ai/blog/articles/how-to-serve-sage-a1-5-sglang-vllm) (2026-10-23): Serve Sage-A1.5 with SGLang or vLLM: standard and MTP launch commands, parsers, and the YaRN factor 2.0 setup that extends 256K context to 512K.
- [Does Vast-10M's 10-Million-Token Context Window Work in Practice?](https://goremy.ai/blog/articles/vast-10m-10-million-token-context-review) (2026-10-23): Hands-on look at Vast-10M, which swaps attention in DeepSeek V4 and GLM 5.2 for VSA. Results from a game build, HTML art, a legal corpus and 81 languages.
- [Sage-A1.5 Benchmarks: How Good Is Shanghai AI Lab's 35B Agent Model?](https://goremy.ai/blog/articles/sage-a1-5-shanghai-ai-lab-agentic-model) (2026-10-23): Sage-A1.5 is a 35B MoE agentic model from Shanghai AI Lab. See GAIA, BrowseComp and SWE-bench Pro scores against larger models, plus SGLang and vLLM serving.
- [What is Raon-1 mini and can it run on one RTX 5090?](https://goremy.ai/blog/articles/raon-1-mini-preview-krafton-open-model) (2026-10-23): Raon-1 mini preview is KRAFTON AI's Apache 2.0 30B-A4B MoE with hybrid attention and 262K context. Specs, NVFP4 on a 32GB RTX 5090, and benchmarks.
- [Can You Run Fish Audio Drama 3 Locally? VRAM and Setup Guide](https://goremy.ai/blog/articles/fish-audio-drama-3-run-locally-vram) (2026-10-23): Fish Audio Drama 3 is an open-weight MoE text-to-speech model. Here are the VRAM needs, the 4-bit quant, bracket directions, and how it ranks.
- [Why Does Temperature 0 Still Give Different Answers in LLMs?](https://goremy.ai/blog/articles/temperature-zero-different-answers-batch-invariant) (2026-10-22): Temperature 0 isn't always deterministic. Batch composition changes GPU math and flips tokens. Here's why, and when to use batch-invariant inference.
- [How Much RAM and Storage Does a Mac mini Need for AI Agents?](https://goremy.ai/blog/articles/mac-mini-for-ai-agents-how-much-memory) (2026-10-22): Buying a Mac mini to run AI agents all day? Here is how to size RAM and storage, what local models change, and why a bigger machine won't fix agent bloat.
- [How does Ornith-1.5-122B score against Qwen 3.8 and DeepSeek V4 Flash?](https://goremy.ai/blog/articles/ornith-1-5-122b-benchmarks-vs-qwen-deepseek-nemotron) (2026-10-22): Ornith-1.5-122B posts 76.9 on SWE-bench Pro and 84.5 on Verified. Here is how it compares with Qwen3.8, Nemotron 3.5 Super VL, DeepSeek-V4-Flash and Beam.
- [How does Hy-Memory-30B-A3B score against Claude Opus 5?](https://goremy.ai/blog/articles/hy-memory-30b-a3b-benchmark-results) (2026-10-22): Tencent's Hy-Memory-30B-A3B memory writer averages 55.49 across four harnesses vs Claude Opus 5 at 54.62. Per-harness and per-benchmark scores inside.
- [How does Spark-X2.6-4B compare with Qwen3.5-4B and Gemma-E4B?](https://goremy.ai/blog/articles/spark-x2-6-4b-vs-qwen-3-5-4b-gemma-e4b) (2026-10-22): Spark-X2.6-4B claims best-in-class small-model results and efficiency on par with Qwen3.5-4B and Gemma-E4B. What the card says and what it leaves out.
- [How to Run Ornith-1.5-122B Locally with llama.cpp and vLLM](https://goremy.ai/blog/articles/run-ornith-1-5-122b-locally-llama-cpp) (2026-10-22): Run Ornith-1.5-122B locally: llama.cpp GGUF setup, VRAM and context trade-offs, runtime versions to check, and how to handle thinking-mode output.
- [How Good Is DeepSeek-V4.1-Flash-1017 vs Opus 5 and GPT-5.6 Sol?](https://goremy.ai/blog/articles/deepseek-v4-1-flash-1017-benchmarks) (2026-10-22): DeepSeek-V4.1-Flash-1017 gains 11.1 points on Terminal-Bench 4.0 over V4.1-Flash. See how it stacks up against Opus 5, GPT-5.6 Sol, GLM-5.3 and K3.
- [How to Run Hy-Memory-30B-A3B Locally with vLLM for Mem0 and MemOS](https://goremy.ai/blog/articles/run-hy-memory-30b-a3b-vllm-mem0) (2026-10-22): Serve Tencent's Hy-Memory-30B-A3B with vLLM as hy-memory, call it via the OpenAI-compatible API, and use it as the write path in Mem0 or MemOS.
- [How to Run Spark-X2.6-4B Locally: Ollama, llama.cpp, MLX, vLLM](https://goremy.ai/blog/articles/spark-x2-6-4b-on-device-agent-model) (2026-10-22): Run Spark-X2.6-4B, an Apache 2.0 4B on-device agent model with 1M-token context, via Ollama, llama.cpp, LM Studio, MLX, vLLM or SGLang.
- [How to Run Nemotron Dory 1.0 4B Locally and Pick a Loop Count](https://goremy.ai/blog/articles/run-nemotron-dory-4b-locally-loop-count) (2026-10-22): Run NVIDIA Dory1.0-4B locally with vLLM or llama.cpp on Apple Silicon, choose 1, 2 or 4 loops, and cut KV cache memory with last-loop caching.
- [How to Run DeepSeek-V4.1-Flash-1017 Without a Jinja Chat Template](https://goremy.ai/blog/articles/run-deepseek-v4-1-flash-1017-locally) (2026-10-22): DeepSeek-V4.1-Flash-1017 ships MIT-licensed with no Jinja template. See sampling settings, 1M context, the encoding folder and deepseek-recipe.
- [What Is Nemotron 3.5 Super VL? Architecture, Speed and 1M Context](https://goremy.ai/blog/articles/nemotron-3-5-super-vl-architecture-speed-context) (2026-10-21): NVIDIA's open-weight Nemotron 3.5 Super VL uses Mamba-2, MoE and attention with 120B total, 12B active parameters and 1M context. Here is where it fits.
- [How Do You Control an LG or Android TV with Hermes Agent and Ollama?](https://goremy.ai/blog/articles/control-smart-tv-hermes-agent-ollama) (2026-10-21): Enable developer mode on an LG or Android TV, find its IP, and point Hermes agent at a local Qwen model in Ollama to manage settings with an undo list.
- [How Good Is Nemotron 3.5 Super VL for Agentic Coding and Vision?](https://goremy.ai/blog/articles/nemotron-3-5-super-vl-hands-on-test) (2026-10-21): Hands-on results for NVIDIA's open Nemotron 3.5 Super VL (120B, 12B active): bug fixing, vision, translation, circuits, and benchmarks.
- [How Does Ornith-1.5-122B Perform on Real Tasks? Hands-On Test Results](https://goremy.ai/blog/articles/ornith-1-5-122b-hands-on-review) (2026-10-21): Ornith-1.5-122B tested on a Docker bug fix, vision, screenshot reasoning, creative HTML and 80-language translation. Here is where it held up and failed.
- [How to Run Local AI Models: Hardware, Tools, and Costs](https://goremy.ai/blog/articles/how-to-run-local-ai-models-hardware-tools-and-costs) (2026-10-21): Learn how to run local AI models with Ollama or LM Studio on a laptop, GPU, or phone, plus local image and music generation and API cost break-even.
- [How to Run CUDA on AMD GPUs With ZLUDA on Windows](https://goremy.ai/blog/articles/how-to-run-cuda-on-amd-gpus-with-zluda-on-windows) (2026-10-21): Learn how to run CUDA apps on AMD GPUs with ZLUDA on Windows: setup steps, supported GPUs, legal risk under NVIDIA's EULA, and what still breaks.
- [How to Run a Local AI Coding Agent on a Mac Mini (M4)](https://goremy.ai/blog/articles/how-to-run-a-local-ai-coding-agent-on-a-mac-mini-m4) (2026-10-21): Build a local AI coding agent on an M4 Mac Mini with Ollama's MLX backend, Qwen3-Coder-Next, Serena MCP tools, and a Panel chat UI. No token billing.
- [What Is an Agentic OS? Windows vs Omarchy Compared](https://goremy.ai/blog/articles/what-is-an-agentic-os-windows-vs-omarchy-compared) (2026-10-21): An agentic OS lets AI agents act on files, settings, and apps. See how Windows 11 and Omarchy Linux differ on who controls the agent, plus real risks.
- [How to Run AI Coding Agents Locally with Magnitude](https://goremy.ai/blog/articles/how-to-run-ai-coding-agents-locally-with-magnitude) (2026-10-21): Learn how to run Claude Code, Codex, or Cline locally with Magnitude in three steps, plus DeepSeek vs OpenAI cost math and the tool's alpha limits.
- [How to Run Distilled AI Models Locally (DeepSeek-R1, Ollama)](https://goremy.ai/blog/articles/how-to-run-distilled-ai-models-locally-deepseek-r1-ollama) (2026-10-21): Learn how to pick, quantize, and run distilled models like DeepSeek-R1 locally with Ollama or llama.cpp, and when self-hosting beats API costs.
- [Is Omarchy Linux Good for Developers? Pros, Cons, and Who It Fits](https://goremy.ai/blog/articles/is-omarchy-linux-good-for-developers-pros-cons-and-who-it-fits) (2026-10-21): Omarchy is DHH's Arch and Hyprland desktop with built-in AI coding agents. See review scores, security issues, costs vs Codespaces, and who should switch.
- [Best Tools for Running AI Locally: Ollama vs LM Studio vs vLLM](https://goremy.ai/blog/articles/best-tools-for-running-ai-locally-ollama-vs-lm-studio-vs-vllm) (2026-10-21): Compare Ollama, LM Studio, llama.cpp, Open WebUI, AnythingLLM, and vLLM across runtime, interface, and server layers, plus VRAM needs and cost.
- [Best Open Source AI Models for Internal Tools: Which to Pick?](https://goremy.ai/blog/articles/best-open-source-ai-models-for-internal-tools-which-to-pick) (2026-10-21): Compare Llama, Qwen, Mistral, DeepSeek, gpt-oss, and small models by license, hardware needs, and self-hosting break-even to pick one for internal tools.
- [How to Build a Local AI Second Brain: Step-by-Step Guide](https://goremy.ai/blog/articles/how-to-build-a-local-ai-second-brain-step-by-step-guide) (2026-10-21): Build a local AI second brain with markdown notes, Ollama, a local vector store for RAG, and automation. Covers hardware needs, tradeoffs, and maintenance.
- [How to Run DeepSeek R1 671B Locally by Streaming Experts Off SSD](https://goremy.ai/blog/articles/how-to-run-deepseek-r1-671b-locally-by-streaming-experts-off-ssd) (2026-10-21): Learn how to run 671B MoE models like DeepSeek R1 locally with llama.cpp, KTransformers and Unsloth quants, with hardware tiers and tokens per second.
- [Is a Local LLM Cheaper Than an API for Business? Break-Even Math](https://goremy.ai/blog/articles/is-a-local-llm-cheaper-than-an-api-for-business-break-even-math) (2026-10-20): Learn how to run a local LLM for business: hardware tiers, 7B-30B model picks, Ollama and vLLM setup, and when owning GPUs beats paying API bills.
- [What Is an LLM Proxy? How to Set One Up for Local AI Agents](https://goremy.ai/blog/articles/what-is-an-llm-proxy-how-to-set-one-up-for-local-ai-agents) (2026-10-20): Learn what an LLM proxy is and how to set one up with LiteLLM to log prompts, tool calls, and spend from local AI agents like Ollama, Cline, and Continue.
- [Is Local AI Cheaper Than APIs? DeepSeek Ban and Self-Hosting Costs](https://goremy.ai/blog/articles/is-local-ai-cheaper-than-apis-deepseek-ban-and-self-hosting-costs) (2026-10-20): DHS banned DeepSeek from 100+ federal agencies. See what running AI models locally costs, from a $6,000 CPU build to H100 clusters, and when it beats APIs.
- [Open Source AI vs SaaS AI: The Privacy Case for Owning Your Models](https://goremy.ai/blog/articles/open-source-ai-vs-saas-ai-the-real-privacy-case) (2026-10-19): Court orders, training opt-outs, and leaks at OpenAI, Anthropic, and xAI show renting AI exposes your data. Here's the case for running open weights yourself.
- [Omarchy Linux Review: AI Capabilities, Tested Hands-On](https://goremy.ai/blog/articles/omarchy-linux-review-ai-capabilities-hands-on) (2026-10-15): Omarchy ships 10 AI coding agents built into the desktop. A hands-on look at what works, what's shaky, and who should actually run it.
- [The API Tax: How to Distill Frontier Models for Enterprise](https://goremy.ai/blog/articles/how-to-distill-frontier-models-for-enterprise) (2026-10-09): Enterprises spend $12.5B a year renting frontier AI APIs. Here's the distillation pipeline that turns that spend into owned models, 75-98% cheaper to run.
- [GPU Passthrough on KVM: The Setup Guide for Local AI Coding Agents](https://goremy.ai/blog/articles/gpu-passthrough-on-kvm-local-ai-coding-agent-setup-guide) (2026-10-05): Step-by-step KVM/VFIO GPU passthrough guide: IOMMU, VFIO binding, libvirt/Proxmox config, and model serving for near-native local AI coding agents.
- [How to Run AI Models Locally in 2026](https://goremy.ai/blog/articles/how-to-run-ai-models-locally-in-2026-a-practical-guide) (2026-10-04): GLM-5.3-Flash and 512GB unified-memory hardware make local AI practical in 2026. Here's the setup, the real cost math, and when renting still wins.
- [How to Run AI Agents Locally on Your Own Hardware](https://goremy.ai/blog/articles/how-to-run-ai-agents-locally-a-practical-setup-guide) (2026-10-04): Open-weight models now rival GPT-4o and o1. Here's how to run an AI agent locally: hardware, quantization, MCP tool access, and the real cost math.
- [The 5 Best Small AI Models for Local Inference in 2026](https://goremy.ai/blog/articles/5-best-small-ai-models-for-local-inference-in-2026) (2026-10-01): Jev, Julia-1, Ember-1, Bonsai 2, and OrcaSAQ2: five tiny, permissive models that run classification and reasoning on local hardware for near-zero cost.
- [How to Run a Local LLM on Old Hardware: The 2026 Playbook](https://goremy.ai/blog/articles/how-to-run-a-local-llm-on-old-hardware-2026-guide) (2026-09-30): Quantized GGUF models, BitNet 1-bit weights, and P2P clusters like exo and Petals let you run real AI on a Raspberry Pi or old laptop. Here's the playbook.
- [Run Local AI Agents on Quantized Models: OrcaSAQ2 vs. Bonsai 2 Tested](https://goremy.ai/blog/articles/run-local-ai-agents-quantized-orcasaq2-and-bonsai-2-tested) (2026-09-28): OrcaSAQ2 27B and Bonsai 2 27B run agent loops on a laptop for near-zero cost. Here's what their own benchmarks say about the agentic accuracy tradeoff.
- [Omarchy: How to Manage the Employee AI Agents You Can't Ban](https://goremy.ai/blog/articles/how-to-manage-employee-ai-agents-the-omarchy-playbook) (2026-09-26): Pizza Bot and Grok Bot made employee-built AI agents trivial to deploy. Here's how to inventory, own, and govern them instead of trying to ban them.
- [What Is Omarchy? Inside the 'AI Operating System' Debate](https://goremy.ai/blog/articles/omarchy-explained-ai-operating-system-or-just-dotfiles) (2026-09-25): Omarchy pre-wires ten AI agent CLIs into Arch Linux at the OS level. What it actually is, what it isn't, and who controls the config and the money.
- [How to Run Local AI on Old Hardware (No GPU Required)](https://goremy.ai/blog/articles/running-local-ai-on-old-hardware-step-by-step-guide) (2026-09-25): Old laptops can run local AI. See how Omarchy Linux and Ollama turn a decade-old, GPU-less machine into a working agentic coding setup.
- [How to Run Frontier AI on Local Enterprise Hardware](https://goremy.ai/blog/articles/how-to-run-frontier-ai-on-local-enterprise-hardware) (2026-09-24): Open-weight models like gpt-oss-120b now run on enterprise hardware from $2K to $30K. Here's the playbook and the real break-even math against your SaaS bill.
- [Exfiltrate Your Weights: The Push for Mission-Critical Local AI](https://goremy.ai/blog/articles/running-ai-models-locally-vs-api-the-enterprise-case-in-2026) (2026-09-24): Banks, hospitals, and governments are downloading open-weight models and running them on owned infrastructure. Here's the real math on local AI vs API.
- [Run AI Locally or Pay for Cloud SaaS? The 2026 Math](https://goremy.ai/blog/articles/run-ai-locally-vs-cloud-saas-the-real-2026-cost-case) (2026-09-21): Omarchy's $18M backing and 192GB desktops make local AI a real alternative to cloud APIs. Here's the honest cost, privacy, and capability math.
- [The 2026 Self-Hosted AI Development Stack: Rails 8, Omarchy, and Ollama](https://goremy.ai/blog/articles/self-hosted-ai-development-stack-rails-8-omarchy-ollama) (2026-09-12): A self-hosted AI development stack in 2026: Rails 8, Omarchy, and Ollama replace your PaaS, Redis, and API bills. The math, the 37signals case study, and the limits.
- [Buy GPUs or Rent Cloud AI? The Real Break-Even Math](https://goremy.ai/blog/articles/buy-gpus-or-rent-cloud-ai-the-real-break-even-math) (2026-09-11): Is it cheaper to buy GPUs or rent cloud AI for internal tools? The break-even is lower than you think. Here's the real math and a decision framework.
- [The Best Local Open-Source LLM for Coding in 2026](https://goremy.ai/blog/articles/best-local-open-source-llm-for-coding-2026-guide) (2026-09-10): Five Apache-licensed local coding models that match GPT-4o on benchmarks, run on one GPU, and cost nothing per token. A practical buyer's guide.
- [How Developers Reuse Old Hardware with Local AI Scripts](https://goremy.ai/blog/articles/how-to-reuse-old-hardware-with-local-ai-scripts) (2026-09-09): Learn how developers turn rotary phones, old laptops, Kindles, and spare phones into a subscription-free smart home using open-source local AI scripts.
- [Mac Studio vs Cloud APIs: The Real Cost of Running Local AI Agents](https://goremy.ai/blog/articles/mac-studio-vs-cloud-api-costs-for-local-ai-agents) (2026-09-08): Mac Studio TCO vs cloud API subscriptions for local AI agents: the three-year math, the 94/6 routing trick that cuts bills by 90%+, and the hardware catch.
- [Self-Hosted AI vs Cloud AI: What the Triple Outage Proved](https://goremy.ai/blog/articles/self-hosted-ai-vs-cloud-ai-what-the-outage-proved) (2026-09-06): ChatGPT, Claude, and Grok all went down the same morning. Here's what that reveals about self-hosted AI vs cloud AI, and how to de-risk your stack.
- [Local AI vs Cloud LLM for Enterprise: Why the Real Shift Is Happening Quietly](https://goremy.ai/blog/articles/local-ai-vs-cloud-llm-for-enterprise-the-real-2026-tradeoff) (2026-09-06): GPT-6 Astra owns the headlines, but enterprise inference is quietly shifting local. Here's the cost, governance, and workload math behind the move.
- [What Hardware Do You Actually Need to Run LLMs Locally?](https://goremy.ai/blog/articles/hardware-requirements-for-local-llm-inference-2026-guide) (2026-09-03): What GPU, memory, and budget you actually need to run a 27B-class model like Qwen3 locally, plus the break-even math against API pricing.
- [The Best Local Open-Source AI and Infrastructure Tools in 2026](https://goremy.ai/blog/articles/best-local-open-source-ai-infra-tools-2026) (2026-09-02): Best local open-source AI and infrastructure tools in 2026: Ollama, llama.cpp, Valkey, OpenTofu, and n8n for self-hosting.
- [The Mac Studio M5 Ultra Makes the Case for Owning Your AI Compute](https://goremy.ai/blog/articles/mac-studio-m5-ultra-the-case-for-owning-your-ai-compute) (2026-09-02): Local AI vs cloud AI cost, explained with real numbers. Why Apple's M5 Ultra and open models like Granite 4.2 change the local-versus-cloud math.
- [Building a Local AI Agent with a Raspberry Pi and Read-Only Postgres](https://goremy.ai/blog/articles/run-an-llm-locally-on-a-raspberry-pi-with-read-only-postgres) (2026-09-02): A hands-on test of running local LLMs on a Raspberry Pi 5 with a read-only Postgres role, proving cheap edge hardware can power governed internal AI tools.
- [Local AI Models vs API Cost: The Real Break-Even Math](https://goremy.ai/blog/articles/local-ai-models-vs-api-cost-the-real-break-even-math) (2026-08-31): GLM-5.3-Flash and Qwen3.8 now run on a single GPU. Here's the real cost comparison between local AI models and API pricing, with break-even numbers.
- [What Is an AI Inference Engine? vLLM Explained](https://goremy.ai/blog/articles/what-is-an-ai-inference-engine-vllm-explained) (2026-08-31): An AI inference engine like vLLM serves trained model weights to real users. Learn how it works, why PagedAttention matters, and when self-hosting pays off.
- [How to Run GLM-5.3-Flash Locally on CPU (No GPU Required)](https://goremy.ai/blog/articles/run-glm-5-3-flash-locally-on-cpu-no-gpu-needed) (2026-08-31): How to run the 320B-parameter GLM-5.3-Flash model on CPU with 128GB RAM using Unsloth's dynamic GGUF quants, plus the hardware math and exact commands.
- [Reclaiming the Dev Cycle: Building an AI-Native Workflow You Actually Own](https://goremy.ai/blog/articles/ai-native-development-cycle-build-one-you-actually-own) (2026-08-28): Shadow AI is now standard, not fringe. Here's how to build a standardized, locally run AI development cycle with small models and DuckDB instead.
- [When the Cloud Breaks: Why Local AI Is the Ultimate Software Asset](https://goremy.ai/blog/articles/when-the-cloud-breaks-why-local-ai-wins-on-reliability) (2026-08-28): Cloud AI outages keep hitting OpenAI, Anthropic, and AWS. Here's why local AI execution is becoming the safer, cheaper standard for software ownership.
- [Tuning Your Local AI: Why Self-Hosted Models Feel Dumber Than They Are](https://goremy.ai/blog/articles/how-to-make-a-local-llm-smarter-a-practical-tuning-guide) (2026-08-25): Local LLMs feel dumb because of defaults, not weak models. Fix context length, quantization, chat templates, and sampling to close the gap with APIs.
- [How to Self-Host Your Company's AI Stack: Replacing SaaS APIs with Local Models](https://goremy.ai/blog/articles/how-to-self-host-ai-models-for-business-2026-guide) (2026-08-25): A practical howto on self-hosting your company's AI stack: break-even math, open models like DeepSeek V3, and tools like SIE and FreeToken.