BotPsy-0.8B

BotPsy-0.8B is a small language model that selects and calls software tools from text instructions. Tether AI Research’s QVAC team introduced it on 11 November 2026 for low-to-medium difficulty tasks.

A small model can handle a bounded tool task

The model card identifies Qwen3.5-0.8B as the base model. BotPsy is intended for text input, returning text and tool calls. The developers recommend English. Those stated limits are the appropriate basis for use, even where an automatically generated hosting example reflects the broader base architecture.

A tool-calling application supplies the model with available functions and their arguments, then decides whether to execute the proposed call. Examples include choosing a search function or requesting a calculation. The surrounding application remains responsible for permissions and validation before a tool changes data or takes an action.

Downloadable formats support different runtimes

QVAC publishes BF16 safetensors weights and separate GGUF files in BF16, Q8_0 and Q4_K_M formats. Quantised formats trade numerical precision for a smaller memory footprint. Actual speed and memory requirements depend on the runtime, device and request length.

The model uses Apache 2.0. Check the repository’s current files and runtime instructions when choosing a format. The presence of downloadable weights is separate from any hosted service’s pricing or availability.

Evaluate the complete application

The launch benchmarks are developer-run results. For a useful comparison, test tool selection, argument accuracy, failed calls and requests that should require human review. A small model can suit constrained jobs while still needing a larger model or a person for harder cases.

Image: QVAC’s official BotPsy artwork.