Wholesale, resold to you
One balance across 60+ models, bought in volume so the per-token rate drops well below list.
Every frontier model
Official list price per 1 million tokens (input / output), shown next to what you effectively pay through Somnus after the 2× bulk credit.
| Model | Maker | Official input /1M tokens | Official output /1M tokens | Somnus effective (≈ half) |
|---|---|---|---|---|
| Claude Sonnet 4.5 | Anthropic | $3.00 | $15.00 | $7.50 |
| Claude Opus 4.1 | Anthropic | $15.00 | $75.00 | $37.50 |
| GPT-5 | OpenAI | $1.25 | $10.00 | $5.00 |
| GPT-5 mini | OpenAI | $0.25 | $2.00 | $1.00 |
| Gemini 2.5 Pro | $1.25 | $10.00 | $5.00 | |
| Gemini 2.5 Flash | $0.15 | $0.60 | $0.30 | |
| Grok 4 | xAI | $3.00 | $15.00 | $7.50 |
| DeepSeek V3.2 | DeepSeek | $0.14 | $0.28 | $0.14 |
| DeepSeek R1 | DeepSeek | $0.55 | $2.19 | $1.09 |
| Kimi K2 | Moonshot | $0.60 | $2.50 | $1.25 |
| GLM-4.6 | Zhipu | $0.25 | $1.00 | $0.50 |
| Qwen3 Max | Alibaba | $0.50 | $2.00 | $1.00 |
Effective price = official output price ÷ 2, because 2× bulk credit means every dollar buys two dollars of compute. Savings skill can reduce it further.
Somnus vs the subscription apps
The Claude app and Codex / ChatGPT Plus charge a flat $20/month and cap your usage. Somnus sells you the raw model at roughly half the official price — no cap, no throttle — and a savings skill trims another ~10% off every chat.
2× bulk credit halves the official rate; the savings skill takes another ~10% off every chat.
Subscription apps throttle heavy users. Somnus is pure pay-as-you-go — spend your balance however you like.
Claude, GPT, Gemini, Grok, DeepSeek and more behind a single API key and desktop agent.
About Somnus
Somnus is a desktop AI agent with every major model built in — paid for with simple prepaid credit that costs roughly half the official API price.
Chat, tools, terminal, files, memory, skills, and sessions live together. Switch models without changing your setup or juggling accounts.
The Somnus desktop app is based on Hermes Agent by Nous Research, released under the MIT License. The original copyright and permission notice ship with the app.
View upstream projectHow the credit works
We buy large blocks of API capacity from every provider at wholesale rates, then pass the discount straight to your balance. Every top-up is doubled — pay $10, get $20 of real API spend. Your first top-up gets a 2.5× boost. Credit never expires.
One balance across 60+ models, bought in volume so the per-token rate drops well below list.
Put in $10 and you get $20 of compute at the official API price — effectively half price, always.
A built-in skill routes each request to the cheapest model that still does the job, trimming another ~10% off every chat.