🔍 Read the full analysis: BotPsy: On-Device Personal Assistant With A Sub-Billion-Parameter Agentic Language Model on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Tether AI Research has introduced BotPsy, a 0.8-billion-parameter model based on Qwen3.5-0.8B and designed for tool use and agent workflows, including on-device deployment. The company reports improved scores over the base model on several benchmarks, but its results have not been independently verified and do not establish reliability or safety 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, as described in the original analysis. The company says it is designed for low- to medium-difficulty tasks and can run on-device, but its performance figures are company-reported and no independent or real-world deployment results are provided.
BotPsy is presented as a personal assistant that uses tools for tasks including web research, travel and lifestyle requests, commerce and finance, business operations, device controls, documents, calendars and messaging. Tether AI Research says the model is intended for routine work rather than high-stakes decisions, and acknowledges that it can make mistakes.
In the company’s reported evaluations, BotPsy scored 44.0 on BFCL v4, compared with 23.4 for its Qwen3.5-0.8B-Think base model. It scored 58.7 on ACEBench versus 30.1, and 51.5 on τ-bench versus 23.2. Tether says the τ-bench score is the mean across its retail, telecom and banking subsets. The supplied information does not describe an independent test or provide full testing protocols.
The company says BotPsy led models of up to 1.2 billion parameters across six tested tool-use benchmarks. Against the larger Qwen3.5-2B-Think, results were mixed: BotPsy scored higher on BFCL v4, ACEBench, τ-bench, ToolSandbox and BrowseComp-Plus, while Qwen3.5-2B-Think scored 48.9 on Claw-Eval to BotPsy’s 46.5. These are benchmark comparisons reported by the developer, not evidence that BotPsy will outperform larger models on every task.
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.
The Case for Local Assistant Tasks
BotPsy’s small parameter count is relevant because it may make a tool-using assistant easier to run locally than a larger model, if supported hardware and software can run it effectively. On-device processing could allow some routine requests to be handled without sending them to a cloud service. That may appeal to users concerned about data leaving their devices, and to organizations seeking alternatives to relying on external services for every assistant request.
Those potential benefits remain conditional. The supplied material does not establish BotPsy’s memory, power or latency requirements on consumer devices, nor does it provide a verified comparison of operating costs with cloud-based models. A small model can still fail when it selects the wrong tool, misinterprets instructions or mishandles information. Tether’s results make the model a development worth tracking, but benchmark scores alone do not show dependable performance in ordinary use.
The safety figures also require careful interpretation. Tether reports an 86.1 score on BFCL v4 hallucination tests, which it describes as evaluating whether a model abstains when no suitable tool is available. It also reports a 31% unsafe-response rate under prompt-injection attacks, down from 46% for the base model. These figures are limited to the company’s reported tests; they do not establish that the model is safe across different tools, prompts or deployments.
Products Worth Considering
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
DJI Osmo Pocket 3 Vlogging Camera with 1-Inch CMOS Sensor and 4K/120fps Video
Bose Lifestyle Ultra Soundbar and Ultra Subwoofer Bundle
As an affiliate, we earn on qualifying purchases.
A Specialized Qwen3.5-Based Model
BotPsy is based on Qwen3.5-0.8B and is focused on function calling and agentic workflows, rather than being presented as a general-purpose frontier model. The source lists model weights in safetensors and GGUF formats, with bf16, Q8 and Q4 versions. It does not provide enough detail to establish compatibility with particular devices or software runtimes.
Tool-using models differ from chat systems that only generate text: they are expected to choose and call tools, then use the results to respond. Their usefulness depends not just on language ability but also on selecting appropriate actions and declining when available tools cannot satisfy a request. Tether’s benchmark claims address aspects of that capability, but the supplied information does not include user deployment studies or independent reproductions.
The company also reports gains in instruction-following and some knowledge benchmarks, while noting that results weaken as instructions accumulate and in lower-resource languages. The announcement material does not quantify those limitations. Tether’s stated rationale is that smaller specialized models may handle some routine tasks with lower operating costs and more local data control; that cost and privacy case is a proposal, not a verified comparison.
“BotPsy is a “personal assistant” that uses tools to perform a variety of simple tasks for the user.”
— Tether AI Research
Deployment and Evaluation Gaps
The supplied material does not identify an independent evaluation or give full testing protocols, so it is unclear whether the reported scores would be reproduced by other evaluators or implementations. Benchmark performance may not predict reliability when tools change, user requests are ambiguous, or a task involves a long sequence of actions.
There are also practical questions about on-device use: the announcement does not specify supported hardware, memory needs, power consumption, response times or how performance varies between devices. It does not provide a verified end-user cost comparison with cloud models. The company notes weaker results as instructions accumulate and in lower-resource languages, but the supplied details do not quantify those limits.
The material does not state a release date, explain support or update commitments, or describe safeguards in enough detail to assess them across settings. The reported prompt-injection rate and abstention score are company-reported test results, not proof of safety in all applications.
Independent Tests and Device Trials
The next evidence that would help assess BotPsy is independent reproduction of its benchmark results, alongside testing on representative phones, computers or other supported devices. Such evaluations could report response times, memory use, energy consumption and tool-call reliability, including how the model behaves when a workflow fails or continues across multiple steps.
Users and developers will also need clear information on compatible runtimes, supported tools, safeguards and how model updates will be delivered. Tether AI Research has not provided a timeline for further releases or evaluations in the supplied material. Until those details and deployment results are available, BotPsy’s performance beyond the reported tests remains an open question.
Key Questions
What is BotPsy?
BotPsy is a 0.8-billion-parameter model from Tether AI Research, based on Qwen3.5-0.8B and designed for function calling and agent workflows.
Can BotPsy run on a personal device?
Tether says it can be deployed on-device. The supplied announcement does not specify supported devices, hardware requirements, memory use or performance on particular systems.
How did BotPsy perform on benchmarks?
Tether reports higher scores than the Qwen3.5-0.8B-Think base model on BFCL v4, ACEBench and τ-bench. Its comparisons with Qwen3.5-2B-Think were mixed across six tests, and the reported figures have not been independently verified in the supplied material.
Is BotPsy safe for high-stakes tasks?
Tether says BotPsy is not intended for high-stakes use and warns that it can make mistakes. Its reported safety tests do not establish that the model is safe across all tools and real-world deployments.
When will BotPsy be available?
The supplied material does not specify a release date or describe support and update commitments.
Primary source: Hugging Face · via ThorstenMeyerAI.com
