🔍 Read the full analysis: BotPsy And The Case For On-Device Agentic AI Assistance 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 function calling and agent workflows, including possible on-device use. The company reports gains across several benchmarks, but the supplied material includes no independent evaluations or real-device deployment results.
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 aimed at low- to medium-difficulty tasks and can be deployed on-device, but its reported benchmark results have not been independently verified in the supplied material.
BotPsy is presented as a tool-using personal assistant for tasks such as web research, calendars, messaging, device controls, document handling, commerce and business operations. Rather than focus on frontier-scale general performance, its stated purpose is to select and call tools for routine requests. Tether AI Research cautions that the model can make mistakes and is not intended for high-stakes use.
The company reports a 44.0 score on BFCL v4, compared with 23.4 for the Qwen3.5-0.8B-Think base model. It also reports 58.7 versus 30.1 on ACEBench, and 51.5 versus 23.2 on τ-bench. Tether says the τ-bench figure averages its retail, telecom and banking subsets. These are company-reported scores; the supplied information does not describe an independent replication or provide full test protocols.
Tether says BotPsy led models of up to 1.2 billion parameters on six tested tool-use benchmarks. Against the larger Qwen3.5-2B-Think, results were mixed: BotPsy scored higher on five named tests—BFCL v4, ACEBench, τ-bench, ToolSandbox and BrowseComp-Plus—while Qwen3.5-2B-Think scored 48.9 on Claw-Eval, above BotPsy’s 46.5. The model weights are listed in safetensors and GGUF formats, including bf16, Q8 and Q4 variants.
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.
What Local Agent Work Could Gain
If the model performs reliably on suitable devices, on-device tool use could keep some requests and data local, reducing the need to send routine assistant tasks to a cloud service. That could matter to people seeking more control over where their information is processed, as well as developers weighing cloud-service costs and availability. Those are potential benefits, not demonstrated outcomes: the announcement gives no verified cost comparison or reports from deployed users.
Tool-using assistants must do more than generate plausible text. They need to choose an available action, handle a failed tool call and sometimes decline a request when no suitable tool exists. Tether reports an 86.1 score on BFCL v4 hallucination tests, which it describes as testing abstention 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 describe the company’s tests, not a guarantee of safety across applications or attack conditions.
The proposal therefore matters as a test of whether a small model can handle bounded assistant work locally, not as proof that it can replace larger systems. Practical value will depend on its reliability with real tools, the resources it requires, and the safeguards built around it.
Products Worth Considering
Samsung Galaxy Tab S7 11-inch Android Tablet 128GB Wi‑Fi with S Pen Mystic Black
Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB‑C Charging, Up to 2X More Active Noise Cancelling, Transparency Mode, Adaptive
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
A Smaller Model for Tool Calls
BotPsy is built on Qwen3.5-0.8B, with additional focus on function calling and agentic workflows. Tether’s case for a compact model is that some routine tasks may not require a larger cloud model. A smaller model may be easier to run locally, but the supplied announcement does not establish its memory, power or latency requirements across devices.
The results described cover several tool-use benchmarks and include reported gains in instruction-following and some knowledge tests. The source also says performance weakens as instructions accumulate and in lower-resource languages, without quantifying those limits in the supplied details. These caveats matter because everyday requests can involve multiple steps, changing tools and language varieties that benchmark sets may not capture.
“BotPsy is a “personal assistant” that uses tools to perform a variety of simple tasks for the user.”
— Tether AI Research
Device Testing and Safety Gaps
The supplied material does not identify independent benchmark evaluations or describe full testing protocols, so it is unclear whether other evaluators would reproduce the scores. Benchmark results also do not establish how consistently BotPsy will complete longer workflows with changing tools, unexpected inputs or sensitive information.
There are no reported measurements here for memory use, latency, energy consumption or performance on representative devices. The material also does not give end-user costs, quantify the stated limitations with accumulating instructions and lower-resource languages, or specify a release date, supported runtime details, or a schedule for updates. The reported prompt-injection result should not be read as proof that the model is safe in all settings.
Evidence Needed Before Deployment
The next useful evidence would be independent reproduction of the benchmark results and deployment reports from representative phones, computers or other devices. Such testing could show how the model behaves under practical limits on memory and power, how quickly it responds, and whether tool calls remain dependable through longer tasks.
Users and developers will also need clear documentation on supported runtimes, safeguards and update practices. Tether AI Research has not specified a timeline for further evaluations or releases in the supplied material. Until those details emerge, BotPsy’s benchmark results remain an initial company-reported account of a compact model’s capabilities, rather than proof of dependable everyday performance.
Key Questions
What is BotPsy?
BotPsy is a 0.8-billion-parameter model based on Qwen3.5-0.8B and designed for function calling and agent workflows, according to Tether AI Research.
Can BotPsy run on a device?
Tether says the model can be deployed on-device and lists safetensors and GGUF weight formats. The supplied material does not provide measurements of performance or resource use on particular devices.
How strong are its benchmark results?
Tether reports higher scores than its Qwen3.5-0.8B-Think base model on several tool-use tests, including BFCL v4, ACEBench and τ-bench. The figures are company-reported; independent reproduction is not included in the supplied information.
Is BotPsy safe for high-stakes tasks?
No such suitability is established. Tether says the model can make mistakes and is not intended for high-stakes use. Its reported safety tests do not show that it is safe in every setting.
When will BotPsy be available?
The supplied announcement does not state a release date or describe support and update commitments.
Primary source: Hugging Face · via ThorstenMeyerAI.com
