Stop choosing models.Start building.
Meet phantom-1. Intelligent model routing that makes the call for you. One API. Zero guesswork.
- 431 models. One balance.
- A budget and spend-rate cap per key.
- No prompts or responses stored by Phantom.
The right model for the task.
One model name. Three models behind it. Send the first request of a conversation to phantom-1 and it chooses a model for the conversation. You pay the rate of the model that runs.
| Model | Best for | Input / 1M tokens | Output / 1M tokens |
|---|---|---|---|
Nemotron 3.5 Lightning Free nvidia/nemotron-3.5-lightning-free Formatting, extraction, and short summaries. | Formatting, extraction, and short summaries. | Free | Free |
DeepSeek V4.1 Flash deepseek/deepseek-v4.1-flash Everyday code and longer documents. | Everyday code and longer documents. | $0.07875 | $0.315 |
Claude Sonnet 5.5 anthropic/claude-sonnet-5.5 Complex debugging and multi-step reasoning. | Complex debugging and multi-step reasoning. | $2.10 | $10.50 |
- “Summarize this email in two sentences.”goes to Nemotron 3.5 Lightning Free
- “Write a small Python function that parses a date.”goes to DeepSeek V4.1 Flash
- “Find the race condition between these two webhook workers.”goes to Claude Sonnet 5.5
431 models. One API.
Prefer to choose for yourself? Access text, image, video, and speech models through one API and balance. Switch models by changing the model ID.
- OpenAI
- Anthropic
- xAI
- Meta
- DeepSeek
- Mistral
- Qwen
- Moonshot
- Z.ai
- MiniMax
- NVIDIA
- Amazon
From terminal to inference in minutes.
Use your Phantom balance from the terminal or the API. Run supported coding assistants through Phantom, or connect an OpenAI-compatible client to the same endpoint.
Install the Phantom CLI
npm i -g @phantomcodes/clipnpm add -g @phantomcodes/clibun add -g @phantomcodes/cli- 1
phantom loginSign in, or start a guest workspace. - 2
phantom fund --amount 10 --cardAdd $10 in credit by card. Other supported payment methods are available. - 3
phantom claudeStart Claude Code using your Phantom balance. Codex and pi are supported too.
Using an OpenAI-compatible SDK?
What Phantom keeps, and what you control.
No prompts or responses stored by Phantom.
Phantom records the key, model, token counts and cost of each call, not its text. The privacy page lists everything kept. Model providers have their own policies.
PrivacyA budget and spend-rate cap per key.
Give each key a budget for the month, or for a number of days you choose, and a cap on what it may spend in one minute.
How budgets workA signed receipt for each paid response.
Each paid response carries a signature over the model the provider reported, its token counts and its cost.
How receipts workRates you can see.
Model rates include Phantom's 5% markup. A card purchase adds the card fee at checkout.
Models and rates