Is BotPsy-0.8B state of the art?Pending benchmarks.Announced Nov 11, 2026; no independent results yet.
QVAC (Tether AI Research)'s 0.8B-parameter language model for function calling and agentic workflows, post-trained from Qwen3.5-0.8B for on-device use as a personal assistant. Announced November 11, 2026. Public Apache 2.0 weights are on Hugging Face as BF16 safetensors, with BF16, Q8_0 and Q4_K_M GGUF builds for llama.cpp and the QVAC SDK. No hosted API is offered.
The details so far
- Lab
- QVAC (Tether)
- Announced
- Nov 11, 2026
- API
- Not yet · None; the weights run locally
- Open weights
- Published · Public, ungated Hugging Face repo qvac/BotPsy-0.8B (BF16 safetensors). GGUF builds (BF16, Q8_0, Q4_K_M) in qvac/BotPsy-0.8B-GGUFs
- Parameters
- 0.8B
- Type
- Causal language model for tool calling (Qwen3.5 architecture: hybrid Gated DeltaNet plus gated attention), used text-only, with thinking mode (QVAC model card)
- Base
- Post-trained from Qwen3.5-0.8B in four stages: supervised fine-tuning, multi-teacher on-policy distillation, and two reinforcement-learning rounds (QVAC blog)
- Language
- English is recommended; also evaluated on Multi-IF in seven other languages (QVAC blog)
- License
- Apache License 2.0 (model card and LICENSE file)
- Related
- Separate from VisionPsy-Nano-460M (same lab) and from the Qwen3.5-0.8B base model
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: self-reported BFCL v4 44.0, ACEBench 58.7, and τ-bench 51.5, compared with 23.4, 30.1 and 23.2 for the Qwen3.5-0.8B-Think base (QVAC blog and model card).
Lab claim: self-reported IFEval 90.8 and IFBench 55.9, compared with 59.7 and 22.0 for the Qwen3.5-0.8B-Think base (QVAC blog and model card).
Sources
LABQVAC Hugging Face blog: BotPsyhuggingface.co/blog/qvac/botpsy · Nov 11, 2026REPOqvac/BotPsy-0.8B on Hugging Facehuggingface.co/qvac/BotPsy-0.8BREPOqvac/BotPsy-0.8B-GGUFs on Hugging Facehuggingface.co/qvac/BotPsy-0.8B-GGUFsTracking continues until an independent board lists it.