Praxium
Before your first chat
API setup and pricing
Praxium uses your OpenAI API account for chat. Model choice, research depth, and conversation length affect what OpenAI charges.
Praxium does not charge for any services, including Numidium. Any API usage charges come directly from OpenAI and are paid to OpenAI. None of those payments go to the site’s owner.
Search and the wiki do not require an API key. A ChatGPT subscription does not cover API usage.
Set up your API key
- Open your OpenAI API account. Sign in at OpenAI Platform. Set up API billing in Billing; having a ChatGPT plan alone is not enough.
- Create a dedicated project and key. On the API keys page, create a secret key for Praxium. If you restrict its permissions, allow Responses API requests and access to your chosen model. Keep the key private.
- Save it in your account. Go to Account → OpenAI API key and paste it into the password field. You can replace or remove it there. Never paste a key into a chat message.
- Start small and check usage. Try a short question with a lower-cost model, then review the OpenAI usage dashboard. Set budget alerts in your OpenAI project. Alerts are not a guaranteed spending cutoff.
Praxium stores the key on its server to make requests for your account. Saving it does not verify billing or model access. For an invalid-key error, replace it; for insufficient credits, check API billing; for unavailable models, choose another model or check project access.
Optional Jev discovery
Jev helps Numidium find useful wiki passages, connections, and counterevidence. It is off by default and uses a separate TypeSafe API key.
- Sign in to the TypeSafe Console, open API Keys, create a key, and check your account’s API access and credits.
- Save the key in Praxium Account → Jev. Saving does not enable it or make a paid request.
- Enable Jev in Account or the Chat composer. Linked Discord users can run
/jev enabled:Trueor/jev enabled:False. The setting is shared across web and Discord.
Enabled research turns may send your question, a public-evidence research brief, published wiki passages, and selected source text to TypeSafe. Jev tools exclude Discord messages and private chat history. Each turn permits up to four paid evaluations; a whole-wiki scan may contain thousands of passages. TypeSafe bills your key separately from OpenAI or ChatGPT subscription usage.
TypeSafe’s setup guide · Save your key and choose your setting
Compare model rates
USD per 1 million tokens at OpenAI’s Standard rate. A token is a small piece of text. Input includes your prompt, conversation context, instructions, and retrieved sources.
| Model | Input | Cached input | Cache write | Output |
|---|---|---|---|---|
| GPT-6.1 Sol | $2 | $0.1 | $2.5 | $10 |
| GPT-6 Luna | $0.1 | $0.01 | $0.125 | $0.5 |
| GPT-6 Sol | $2 | $0.2 | $2.5 | $10 |
| GPT-5.6 Luna | $0.2 | $0.02 | $0.25 | $1.2 |
| GPT-5.6 Terra | $2 | $0.2 | $2.5 | $12 |
| GPT-5.6 Sol | $4 | $0.4 | $5 | $20 |
| GPT-6 Astra | $10 | $1 | $12.5 | $50 |
Checked 2026-09-27 against OpenAI pricing. GPT-5.6 Sol’s current rate is promotional, available at least through November 21, 2026. Available models depend on Praxium’s picker and your OpenAI access; this is a reviewed price snapshot.
GPT-6.1 Sol’s rates were checked on September 30, 2026. Numidium defaults to GPT-6.1 Sol with Extra high reasoning on web and Discord; saved personal choices still apply.
Above 272,000 input tokens in one request, these models charge 2× the input and cache rates and 1.5× the output rate for that whole request. Fast mode, regional processing, and other service tiers have different rates. Check OpenAI for current prices.
What thinking effort changes
Effort changes how much reasoning the model does, not the price per token. Hidden reasoning is billed as output. Higher effort can use more tokens and take longer; there is no fixed “high effort” fee or reliable cost per question.
- Low
- A starting point for focused questions and short follow-ups.
- Medium
- A useful starting point for research with several sources.
- High and above
- For difficult comparisons or extended analysis when more time and potential cost are acceptable.
These are starting suggestions, not measured cost guarantees. The picker shows efforts supported by the selected model. See OpenAI’s reasoning guide.
One Praxium answer can require multiple model requests. Source reads, tool results, conversation history, and hidden reasoning all contribute tokens. OpenAI web search adds $0.01 per call plus search-content tokens at model rates when used.
Try a token-cost example
Enter hypothetical totals to compare models. This calculator makes no API calls and does not predict your next answer or enforce a spending limit.
Count each input token once: uncached, cached, or cache-write. Include retrieved search content in input totals. Use the long-context rate only for requests above the threshold; calculate mixed requests separately. Excludes taxes, other tools, and non-Standard service tiers. Your OpenAI bill is authoritative.