Uploaded October 2026 | Updated November 2026, 2 days ago
Run Claude Code on free OpenRouter models, with no graphics card and no Claude subscription. Day 13 of the Free AI Setup series: one free AI tool a day, set up step by step.
Watch the whole series in order: youtube.com/playlist?list=PLfUBzZXRNC-k
Previous, day 12, vLLM on Windows: youtu.be/X63NQmZdCSo
Next, day 14, NVIDIA NIM free API key: youtu.be/Yp9oQYl2KoQ
What you need: Windows 10 or 11, Claude Code installed, and a free OpenRouter API key (day 10 of this series shows how to get one). No GPU, because the models run on OpenRouter's servers. Python for the test app.
In this video:
- Pick a free OpenRouter model that supports tool calling: Cohere North Mini Code (built for coding agents) and Thinking Machines Inkling (975B, about 1M tokens of context)
- Point Claude Code at OpenRouter with PowerShell variables: ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, ANTHROPIC_API_KEY and the model slots
- Why every model slot (Sonnet, Opus, Haiku, subagent) must point to a free model
- First launch, the trust prompt, and the "model not recognized" note
- Check the connection with /status and /model
- Real task: Claude Code builds a Python to-do app, hits a Windows Unicode bug and fixes it itself
- Switch models with /model and run a second task on Inkling
- The real cost (zero) and how many free requests an agent uses
- OpenRouter free limits: 20 requests a minute, 50 a day, or 1,000 a day after buying $10 of credits once
- Save the setup for one project in .claude/settings.local.json, and the login-screen gotcha on a new install
- Tips: an existing Claude login, error 429, and giving small models clear tasks
Chapters
0:00 Claude Code builds an app on a free model, and the plan
0:26 Day 13: Claude Code on OpenRouter free models
0:48 What you need
1:14 Step 1: Choose a free model with tool calling
1:58 Step 2: Project folder and PowerShell variables
3:13 Step 3: Launch Claude Code
3:41 Step 4: Check the connection with /status
4:15 Step 5: First real task, and the workspace gotcha
5:44 Step 6: Switch to a bigger model
6:23 Step 7: The bill, and the free limits
7:24 Step 8: Save the settings for this project
7:47 Two gotchas: the login screen and your key
8:25 Three tips
8:55 Claude Pro or free models
9:18 Tomorrow: NVIDIA NIM free API key
Links
OpenRouter Claude Code guide: openrouter.ai/docs/cookbook/coding-agents/claude-code-integration
OpenRouter free limits: openrouter.ai/docs/api/reference/limits
OpenRouter models list: openrouter.ai/models?max_price=0
Claude Code environment variables: code.claude.com/docs/en/env-vars
Claude Code settings files: code.claude.com/docs/en/settings
Claude Code with an LLM gateway: code.claude.com/docs/en/llm-gateway-connect
Tomorrow, day 14: a free NVIDIA NIM API key.
New free AI tool every day. Subscribe so you don't miss one: youtube.com/@KGPTalkie?sub_confirmation=1
Run Claude Code on free OpenRouter models, with no graphics card and no Claude subscription. Day 13 of the Free AI Setup series: one free AI tool a day, set up step by step.
Watch the whole series in order: youtube.com/playlist?list=PLfUBzZXRNC-k
Previous, day 12, vLLM on Windows: youtu.be/X63NQmZdCSo
Next, day 14, NVIDIA NIM free API key: youtu.be/Yp9oQYl2KoQ
What you need: Windows 10 or 11, Claude Code installed, and a free OpenRouter API key (day 10 of this series shows how to get one). No GPU, because the models run on OpenRouter's servers. Python for the test app.
In this video:
- Pick a free OpenRouter model that supports tool calling: Cohere North Mini Code (built for coding agents) and Thinking Machines Inkling (975B, about 1M tokens of context)
- Point Claude Code at OpenRouter with PowerShell variables: ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN, ANTHROPIC_API_KEY and the model slots
- Why every model slot (Sonnet, Opus, Haiku, subagent) must point to a free model
- First launch, the trust prompt, and the "model not recognized" note
- Check the connection with /status and /model
- Real task: Claude Code builds a Python to-do app, hits a Windows Unicode bug and fixes it itself
- Switch models with /model and run a second task on Inkling
- The real cost (zero) and how many free requests an agent uses
- OpenRouter free limits: 20 requests a minute, 50 a day, or 1,000 a day after buying $10 of credits once
- Save the setup for one project in .claude/settings.local.json, and the login-screen gotcha on a new install
- Tips: an existing Claude login, error 429, and giving small models clear tasks
Chapters
0:00 Claude Code builds an app on a free model, and the plan
0:26 Day 13: Claude Code on OpenRouter free models
0:48 What you need
1:14 Step 1: Choose a free model with tool calling
1:58 Step 2: Project folder and PowerShell variables
3:13 Step 3: Launch Claude Code
3:41 Step 4: Check the connection with /status
4:15 Step 5: First real task, and the workspace gotcha
5:44 Step 6: Switch to a bigger model
6:23 Step 7: The bill, and the free limits
7:24 Step 8: Save the settings for this project
7:47 Two gotchas: the login screen and your key
8:25 Three tips
8:55 Claude Pro or free models
9:18 Tomorrow: NVIDIA NIM free API key
Links
OpenRouter Claude Code guide: openrouter.ai/docs/cookbook/coding-agents/claude-code-integration
OpenRouter free limits: openrouter.ai/docs/api/reference/limits
OpenRouter models list: openrouter.ai/models?max_price=0
Claude Code environment variables: code.claude.com/docs/en/env-vars
Claude Code settings files: code.claude.com/docs/en/settings
Claude Code with an LLM gateway: code.claude.com/docs/en/llm-gateway-connect
Tomorrow, day 14: a free NVIDIA NIM API key.
New free AI tool every day. Subscribe so you don't miss one: youtube.com/@KGPTalkie?sub_confirmation=1










