Run Claude Code for less with onomeo
Claude Code is a coding agent that works in the terminal, and it only speaks the Anthropic format. This site has an endpoint in that format, with two low-cost ways to run it. Accounts that have topped up can use the paid models on the current list at their published prices. In the historical test on 30 September 2026, deepseek-flash cost about $0.005 for reading two files and changing two places. Or connect a provider key of your own on the My providers page, such as the free key NVIDIA offers at build.nvidia.com: the site converts each request, sends it to your provider and charges nothing for it. This site's free models do not work here.
Anthropic's coding agent, on the DeepSeek coding line or your own provider keys.
What to fill in
Anthropic format
Paid site models require a top-up. Free site models are not supported on this endpoint; connected providers can also be used.
Step by step
Screenshots and dated measurements show earlier tests. Their model names and costs are historical; use the current model and pricing shown above.
Pick the model
If your account has topped up, the model name is MODEL_ID and you can skip this step. To use your own keys, connect a key from at least one provider under My providers on the My account page. If you have none, NVIDIA gives out free keys at build.nvidia.com. Once connected, it lists each model under its full name, which starts with my/, with a button to copy it; you need it in the next step.
Open My providers
Once connected, the page lists the provider's models under their full names; press Copy. Save the configuration
On Windows, press the Windows key, type PowerShell and press Enter. On macOS, open Terminal. Press Copy, paste the command in and press Enter. It asks for the model name first: paste the name you copied in the last step and press Enter.
$f = "$HOME\.claude\onomeo.json" if (-not (Test-Path $f)) { $m = Read-Host "Model name (MODEL_ID if you have topped up; for your own key, the full name starting with my/)" New-Item -ItemType Directory -Force -Path (Split-Path $f) | Out-Null [IO.File]::WriteAllText($f, @" { "env": { "ANTHROPIC_BASE_URL": "https://onomeo.com", "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY", "ANTHROPIC_MODEL": "$m", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "$m", "ANTHROPIC_DEFAULT_SONNET_MODEL": "$m", "ANTHROPIC_DEFAULT_OPUS_MODEL": "$m", "CLAUDE_CODE_SUBAGENT_MODEL": "$m" } } "@) Write-Host "Saved: $f" -ForegroundColor Green } else { Write-Host "Config file already exists, left unchanged: $f" -ForegroundColor Yellow }Replace YOUR_API_KEY in the command with your key. Once you create a key in this browser, it is filled in here.
Once the model name is in, a green Saved means the configuration is in place. This configuration is kept apart in onomeo.json and leaves your existing Claude Code settings alone. If the yellow line "Config file already exists, left unchanged" appears, it was written before: to change the model or the key, open the file it names in Notepad.
Start Claude Code with it
Go to your project folder and start Claude Code with the command below. Start it this way whenever you want to use this site.
claude --settings "$HOME\.claude\onomeo.json"
The model name you entered shows under the heading. On start it may first say that it does not recognise the model name and will assume a 200k-token context; that is expected and does no harm.
Check it reaches this site
In Claude Code, type /status and press Enter.
Anthropic base URL reading https://onomeo.com means requests go to this site. Give it a task
Press Esc to close the status page, type what you want done in the input box and press Enter. It asks before it creates or changes a file or runs a command; choose 1. Yes to agree.
Here it was asked to create hello.py: once you agreed, it wrote the file. Measured on 28 September 2026: with NVIDIA's nvidia/nemotron-3-super-120b-a12b, the file above was written in a few seconds; the same day NVIDIA answered deepseek-ai/deepseek-v4.1-flash so slowly that Claude Code kept retrying. When a model is slow, put another name in the file from step 2.
If it will not connect
- The address is https://onomeo.com and nothing more. Do not add /v1 on the end; Claude Code adds the path itself.
- This site's free models come back with the own_key_only error: Claude Code sends the whole conversation and every file it has read with each request, which the free allowance this site shares among everyone cannot carry. Use MODEL_ID (for accounts that have topped up) or a model on your own key, whose name starts with my/.
- On MODEL_ID, the balance is charged what DeepSeek actually charges. Measured on 30 September 2026, reading two files and changing two places took about a minute and about $0.005. Claude, GPT and the other big models cost more per request, and every Claude Code request carries a long set of instructions, so the balance goes fast; for some of them a single request is over the daily big-model limit. They are not a good fit for Claude Code.
- "This model is not among the models of a provider you connected": the name is not on your provider's list; copy the full name again from My providers.
- Claude Code reads and writes files entirely through tool calls, so the model you choose must support them and handle them reliably. If the model says it changed a file and the file is unchanged, move to a stronger model.
- On your own key, this site does not charge for the calls; whatever they cost at the provider is settled between you and them.
- Anthropic does not support connecting Claude Code to models other than Claude. If a Claude Code update stops working with this site, tell us through the contact page.