Apertus v1.5 8B Thinking: reasoning or tool calling — not both

The model page states this plainly: this version doesn't call tools

publicai/apertus-v1.5-8b-thinking's catalog description carries a clear warning: 'this variant does not support function calling; use Apertus v1.5 8B for tool-calling flows.' That means, in the Apertus family, reasoning and tool calling don't combine on the same row — you need to choose between them; there's no model here carrying both at once, unlike the Muse Spark or GLM families.

In exchange, the Thinking version is a reasoning model that works through problems step by step before answering and returns that reasoning trace in a separate field; it's more accurate than the standard 8B version on mathematics, logic puzzles, multi-step analysis, and tasks requiring careful reading.

Which task needs which version

If a task only needs to analyze text input and reason (solving a math problem, explaining a logic puzzle, carefully reading a long text and drawing an inference) and doesn't need to call a tool, the Thinking version is the right choice. If, on the other hand, an agent flow needs to run a database query, call an API, or determine the next step based on a function result, the Thinking version can't cover that task — you need to move to standard Apertus v1.5 8B.

If you want to build a single request flow covering both needs (reasoning first, then calling a tool), that can't be done with one model in this family; instead you'd need a two-stage architecture (produce a plan/analysis with the Thinking version, then hand the result to the standard version or another tool-calling model to make the call).

• Analysis/reasoning only, no tool calling: the Thinking version.
• Tool calling required: standard Apertus v1.5 8B.
• Both needed: you need a two-stage architecture, one model isn't enough.

Turning reasoning off shortens the answer but doesn't add tool calling

You can turn reasoning off by appending the `:fast` suffix to the model id or sending `reasoning: false` in the request body; this noticeably shortens responses and consumes fewer output tokens. But it does NOT turn on function-calling support — turning reasoning off only affects reasoning behavior, the capability list doesn't change.

Frequently asked questions

If I send reasoning: false on the Thinking version, can I call tools?

No, turning reasoning off only skips the reasoning step; function-calling capability is never present on this variant. Use standard Apertus v1.5 8B for tool calling.

How are images sent?

Only as a base64-encoded data URL; sending a remote image URL isn't supported.

Related posts

• publicai/apertus-v1.5-8b-thinking
• Model catalog