TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Tether AI Research has introduced BotPsy, a 0.8-billion-parameter model built on Qwen3.5-0.8B for tool use and agent workflows, including on-device deployment. The company reports strong results across six benchmarks, but the figures are its own evaluations and do not establish how the model will perform in everyday use.
Tether AI Research has introduced BotPsy, a 0.8-billion-parameter model built on Qwen3.5-0.8B for function calling and agent workflows. The company says the model is intended for low- to medium-difficulty tasks and can be deployed on-device, where it could reduce reliance on cloud services for routine assistant work.
BotPsy is presented as a personal assistant that uses tools for tasks such as web research, travel and lifestyle requests, commerce and finance, business operations, device controls, document handling, calendars, and messaging. Tether AI Research says the small model is not intended for high-stakes use and acknowledges that it can make mistakes.
In company-reported results, BotPsy scored 44.0 on BFCL v4, compared with 23.4 for its Qwen3.5-0.8B-Think base; it scored 58.7 on ACEBench versus 30.1, and 51.5 on τ-bench versus 23.2. Tether says the τ-bench figure is the mean of its retail, telecom, and banking subsets. The company reports that BotPsy led models up to 1.2 billion parameters on six tested tool-use benchmarks.
Comparisons with a larger Qwen3.5-2B-Think were mixed across the six tests. BotPsy scored higher on five, including BFCL v4, ACEBench, τ-bench, ToolSandbox, and BrowseComp-Plus. Qwen3.5-2B-Think scored higher on Claw-Eval, with 48.9 to BotPsy’s 46.5. The source also reports gains in instruction-following and some knowledge benchmarks, while noting weaker results as instructions accumulate and in lower-resource languages.
BotPsy brings tool use to a compact model
Tether AI Research says its small assistant is designed for routine agent workflows and on-device use. Its benchmark results are developer-reported and do not establish everyday reliability.
Built for routine assistant work
The model is tuned for function calling and agent workflows, rather than frontier-scale general performance.
Find and gather information
Example uses include web research and travel requests.
Handle everyday workflows
It is presented for commerce, business operations, documents, calendars, and messaging.
Run on-device
Local deployment could reduce reliance on cloud services for routine tasks.
Reported results improved over its base
Tether AI Research reports higher scores for BotPsy on three named benchmarks.
| Benchmark | BotPsy | Qwen3.5-0.8B-Think |
|---|---|---|
| BFCL v4 | 44.0 | 23.4 |
| ACEBench | 58.7 | 30.1 |
| τ-bench | 51.5 | 23.2 |
The larger-model comparison was mixed
BotPsy scored higher on five of six tool-use tests against Qwen3.5-2B-Think.
- BFCL v4
- ACEBench
- τ-bench
- ToolSandbox
- BrowseComp-Plus
- Qwen3.5-2B-Think scored 48.9 on Claw-Eval, compared with BotPsy’s 46.5.
Limits matter for deployment
The supplied material describes areas where performance may fall short and leaves practical questions open.
- Results come from the developer; independent reproduction is not provided.
- Performance drops as instructions accumulate and lags in lower-resource languages.
- Hardware needs, memory use, power use, and real-world reliability are not established.
- Tether says BotPsy can make mistakes and is not intended for high-stakes use.
Small Models Move Assistant Work On-Device
If BotPsy works as intended in practical deployments, its size could make tool-using assistants easier to run locally, potentially keeping some requests and data on a user’s device. Tether AI Research frames this as an alternative for routine work that does not require a frontier-scale cloud model, where dependence on external services can bring variable costs and provider control over availability.
The benchmark results matter because tool-using models must select actions and sometimes decline requests when available tools cannot fulfill them. Tether reports an 86.1 score on BFCL v4 hallucination tests, which assess whether a model abstains when no suitable tool exists. It also reports a 31% unsafe-response rate under prompt-injection attacks, down from 46% for the base model. These are company-reported test results, not evidence that the model is safe in all settings.
Products Worth Considering
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Twilight Blue
Amazon Kindle Paperwhite (2024, 16 GB) – 7-inch Glare-Free Display, Black
Bose Lifestyle Ultra Compact Wireless Smart Home Speaker Bundle, Black
As an affiliate, we earn on qualifying purchases.
Built From Qwen3.5-0.8B
BotPsy is based on Qwen3.5-0.8B and is distributed as model weights, with the source listing safetensors and GGUF formats in bf16, Q8, and Q4. Its stated focus is function calling and agentic workflows rather than general-purpose frontier performance.
The announcement argues that specialized, smaller models can handle some routine tasks at lower operating cost and with more local data control than cloud systems. That is the developer’s rationale for the model, not a verified cost comparison. The supplied material gives benchmark tables and selected figures but no independent evaluation or deployment results from users.
“BotPsy is a “personal assistant” that uses tools to perform a variety of simple tasks for the user.”
— Tether AI Research
Benchmark Claims Await Wider Testing
The reported scores come from Tether AI Research; the supplied material does not identify an independent evaluation, provide full testing protocols, or establish whether results will reproduce across implementations. Performance on benchmarks also may not predict reliability in an actual assistant with changing tools, varied user requests, and sensitive data.
It is not clear how BotPsy’s on-device performance varies by hardware, how much memory or power it requires, or what end-user costs would be compared with cloud models. The source says performance drops as instructions accumulate and lags in lower-resource languages, but does not quantify those limits in the provided details. The announcement also does not specify a release date or describe support and update commitments.
Deployment Evidence Will Matter
The next useful evidence would include independent reproduction of the benchmark results and reports from deployments on representative phones, computers, or other devices. Those tests could show latency, memory use, energy consumption, tool-call reliability, and how the model handles failures over longer workflows.
Users considering BotPsy for practical tasks will also need clear documentation on supported runtimes, safeguards, and update practices. Tether AI Research cautions against high-stakes use; the announcement does not provide a timeline for further releases or evaluations.
Where I land
I see BotPsy as a meaningful attempt to bring tool use to a model small enough for on-device deployment. The reported improvements over its base model, particularly on tool selection and instruction following, make the announcement worth tracking. For people who want routine assistant tasks to run locally, the potential benefit is concrete: less need to send every request to a cloud provider.
The strongest counterargument is that benchmark leadership does not show that a compact agent will reliably complete messy, real-world tasks or protect users from harmful tool outputs. The reported scores and safety rates come from the developer, and the source itself notes limits in longer instruction sequences and lower-resource languages. I would raise my assessment if independent teams reproduced the results and deployment reports showed dependable performance across devices and real workflows; evidence of frequent errors or costly hardware requirements would lower it.
Key Questions
What is BotPsy?
BotPsy is a 0.8-billion-parameter model from Tether AI Research, built on Qwen3.5-0.8B and tuned for function calling and agent workflows.
What tasks is it designed to handle?
The company describes tasks including web research, travel, commerce, business operations, device controls, files, calendars, and messaging. It says the model is best suited to low- and medium-difficulty work.
How does it compare with larger models?
In Tether’s reported tests, BotPsy scored higher than Qwen3.5-2B-Think on five of six tool-use benchmarks. The larger model led on Claw-Eval; the figures have not been independently verified in the supplied material.
Is BotPsy suitable for high-stakes decisions?
No. Tether AI Research says the model can make mistakes and is not intended for high-stakes use.
Source: Hugging Face
