Basic Data
API Configuration
Available API Types
Anthropic Messages API: the API requires a maximum output length, so set an Output Token Limit (without it 16000 is sent, 4096 for the Claude 3 generation). Models released after Claude Opus 4.6 (including Opus 4.7/4.8/5, Sonnet 5, Haiku 5, Fable, Mythos) reject configured Temperature and Top P values, so VARIOS AI omits both for these models. Older models receive Temperature, or Top P only if Temperature is not enabled. The Reasoning Effort maps to Anthropic’s effort levels (Minimal and Low →
low) on Opus and Sonnet 4.6 and later as well as on Fable and Mythos; other models (e.g. Haiku 4.5, Sonnet 4.5, Claude 3.x) get a thinking budget scaled by the effort instead, and Temperature and Top P are not sent while it is active. The thinking budget stays at least 1024 tokens below the Output Token Limit; if the limit is too small for that, thinking stays off. Claude Sonnet 5, Opus 5, Fable and Mythos think adaptively even when the reasoning effort is off and show their reasoning summary in the chat; other models only think when the effort is enabled. Only the plain model endpoint is used: Anthropic-hosted tools such as code execution, web search or MCP are not available, while the local code interpreter, knowledge bases and connectors work as with every other API type. A chat model with this API type cannot be called directly via the Chat Completions API; use the assistant ID of an assistant that uses this model instead.Costs and Limits
Each pricing tier consists of a Context threshold (tokens) and a price. The threshold refers to the prompt size of a request, which is the sum of non-cached input tokens and the tokens read from and written to the cache. Once the prompt reaches the threshold of a tier, that tier’s price applies to the whole request. This affects all four prices, including output tokens. The lowest tier also applies to prompts below its threshold. Without an input or output tier, VARIOS AI does not charge for these tokens. Without an input tier, cache reads and writes without their own tier are not charged either.
Example: With a tier starting at 272000 tokens, a request with 300000 prompt tokens is billed entirely at the price of that tier, not just the 28000 tokens above the threshold.
VARIOS AI does not request prompt caching from Anthropic itself, so cache tokens only arise if you request caching in the Additional parameters field. Anthropic charges twice the input price for its 1-hour cache. VARIOS AI bills these tokens at the cache write price as well. If you use the 1-hour cache, the calculated costs are therefore lower than Anthropic’s invoice.
Behavior Settings
Streaming
User Identification and MIME Types
Check Service Status
DLP Security Settings
Basic DLP Settings
GDPR Protection
Detects personal data in the chat and encrypts it before transfer to the AI.Sensitive File Protection
Blocks the upload of documents with confidential classification.Enterprise DLP Settings
DLP Interface (ICAP)
Sends texts and files for review to the configured ICAP server.Test connection
Use Test connection at the bottom of the Configuration tab to check whether the model is reachable with the current input, even before you save it. The test uses the values in the form, not the saved ones, and does not save anything itself. For this, VARIOS AI sends the message “Ping” to the model and limits the response to 32 tokens. If Supports Streaming is enabled, a second request in streaming mode follows. Both must succeed. The Connection test window shows the result:The provider bills the test requests like regular requests. No costs appear for them in the VARIOS AI dashboard. The connection test is available for chat models, not for embedding and voice models.
References
The References tab shows where the model is used:
Clicking an entry opens the group or the assistant. Personal assistants are not listed individually; their number appears below the list, for example “3 personal assistants”. For a model that has not been saved yet, the tab stays empty.