Run Claude Code for Free on Windows 11 (2026)

The Max Plan is $100 a month. But there's a way to run Claude Code against a local model — no API key, no proxy, no cloud. Ollama now speaks Anthropic's API natively, which means Claude Code in VS Code can talk directly to Gemma 4 running on your own GPU. This video shows the complete setup on Windows 11, tested on an RTX 2060 SUPER with 8 GB of VRAM. We cover how to check your VRAM in Task Manager, pick the right Gemma 4 size for your card, and pull it through Ollama. Then we create a Modelfile to fix the one thing nobody else covers — Ollama's default 4,096-token context window, which causes Claude Code to stop mid-file on coding tasks. We expand that to 32K with three lines. After that, one JSON config file inside the project folder redirects Claude Code from Anthropic's servers to your local model. The video ends with a live toggle showing you how to flip between local and cloud Claude by renaming a single file. All code snippets are in the description below — no typing required. #ClaudeCode #Ollama #Windows11 ──────────────────────────────────────── MODELFILE (save as: Modelfile — capital M, no extension) FROM gemma4:e4b PARAMETER num_ctx 32768 PARAMETER num_predict 4096 ──────────────────────────────────────── SETTINGS.JSON (save to: .claude/settings.json inside your project folder) { "env": { "ANTHROPIC_BASE_URL": "http://localhost:11434", "ANTHROPIC_AUTH_TOKEN": "ollama", "ANTHROPIC_API_KEY": "", "ANTHROPIC_DEFAULT_SONNET_MODEL": "gemma4-claude", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "gemma4-claude", "ANTHROPIC_DEFAULT_OPUS_MODEL": "gemma4-claude", "API_TIMEOUT_MS": "600000", "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1", "CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS": "1", "DISABLE_INTERLEAVED_THINKING": "1", "DISABLE_PROMPT_CACHING": "1" } } For global default (applies to every project), copy the same file to: %USERPROFILE%\.claude\settings.json ──────────────────────────────────────── CHAPTERS 0:00 — Intro: Claude Code offline — still responding 0:37 — Step 1: Check your GPU VRAM in Task Manager 1:09 — Gemma 4: pick the right size for your card 1:54 — Pull Gemma 4 with Ollama (version check first) 2:26 — Why Ollama + Claude Code work without a proxy 2:56 — The Modelfile: fix Ollama's 4K context default 3:25 — Build the gemma4-claude model 4:34 — Verify with ollama list 5:01 — Create .claude/settings.json in VS Code 5:35 — What each config setting does 6:28 — Optional: set Gemma 4 as your global default 7:02 — Reload VS Code + offline sanity check 7:52 — Demo: build a color palette tool with no internet 8:53 — Confirm it's local with ollama ps 9:21 — Toggle: one file rename to switch back to cloud Claude 10:46 — Honesty: when Gemma 4 isn't enough 11:18 — Outro + next video --- LINKS Ollama (download): https://ollama.com Gemma 4 on Ollama: https://ollama.com/library/gemma4 Claude Code VS Code extension: https://marketplace.visualstudio.com/... Claude Code Windows 11 setup guide (watch this first if you haven't installed Claude Code):    • Video   ---