Chinese AI models just got put on notice. OpenAI slashed GPT-5.6 prices. Luna: 80% cheaper. Terra: 20% cheaper. Sol Fast: up to 2.5× faster. Meanwhile, Grok 4.5 High is still on the Pareto frontier

Did you know you can run a private AI locally on many 16GB laptops? No subscription. No API key. After the initial download, it can run offline. When you use a local model and disable Ollama's cloud features, your prompts and responses stay on your device. Here' how. Bookmark this: 1) Install Ollama Mac brew install ollama brew services start ollama Windows winget install --id Ollama.Ollama -e Linux curl -fsSL | sh 2) Run Gemma 4 ollama run gemma4:e4b The default E4B download is about 9.6GB. It can run on many 16GB laptops, but close memory-heavy apps first. Performance depends on your available RAM, GPU and operating system. If it's too heavy, try: ollama run gemma4:e2b No API key. No Python. No CUDA setup is needed for CPU use, although compatible GPU drivers may be required for hardware acceleration. 3) Optional: increase the context Inside the Ollama chat: /set parameter num_ctx 8192 /save gemma4:e4b-8k Then run the saved version: ollama run gemma4:e4b-8k On a 16GB laptop, start with 8192. You can try 16384, but 32768 may cause heavy memory use, swapping or an out-of-memory error. More context = more RAM or VRAM. 4) Optional: enforce local-only mode Create this file: Mac/Linux ~/.ollama/server.json Windows %USERPROFILE%\.ollama\server.json Add: { "disable_ollama_cloud": true } Then restart Ollama. This disables Ollama's cloud models and web-search features. Gemma 4 also accepts images, so it can analyze notes, screenshots, diagrams and error messages on-device. Once the model is downloaded, local prompts have $0 in per-prompt API fees. You still pay for your own hardware and electricity, but there's no token bill.

Did you know you can run a private AI locally on many 16GB laptops? No subscription. No API key. After the initial download, it can run offline. When you use a local model and disable Ollama's cloud features, your prompts and responses stay on your device. Here' how. Bookmark this: 1) Install Ollama Mac brew install ollama brew services start ollama Windows winget install --id Ollama.Ollama -e Linux curl -fsSL | sh 2) Run Gemma 4 ollama run gemma4:e4b The default E4B download is about 9.6GB. It can run on many 16GB laptops, but close memory-heavy apps first. Performance depends on your available RAM, GPU and operating system. If it's too heavy, try: ollama run gemma4:e2b No API key. No Python. No CUDA setup is needed for CPU use, although compatible GPU drivers may be required for hardware acceleration. 3) Optional: increase the context Inside the Ollama chat: /set parameter num_ctx 8192 /save gemma4:e4b-8k Then run the saved version: ollama run gemma4:e4b-8k On a 16GB laptop, start with 8192. You can try 16384, but 32768 may cause heavy memory use, swapping or an out-of-memory error. More context = more RAM or VRAM. 4) Optional: enforce local-only mode Create this file: Mac/Linux ~/.ollama/server.json Windows %USERPROFILE%\.ollama\server.json Add: { "disable_ollama_cloud": true } Then restart Ollama. This disables Ollama's cloud models and web-search features. Gemma 4 also accepts images, so it can analyze notes, screenshots, diagrams and error messages on-device. Once the model is downloaded, local prompts have $0 in per-prompt API fees. You still pay for your own hardware and electricity, but there's no token bill.