iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
FLock.io announced on October 13, 2026, that it launched THIS/THAT Model, an open-source, typed decision model built on Alibaba Cloud’s Qwen3.5-2B. Instead of generating paragraphs, the 1.88-billion-parameter model selects from options defined by a developer, returns probabilities and a confidence score, and is designed to run locally so sensitive inputs can remain on the device.
What THIS/THAT Model does
THIS/THAT accepts a state or prompt, a declared question, and a fixed set of permitted answers. In one forward pass it returns the selected option, its index, probabilities for the available choices, and a confidence value. An application can then route, approve, block, queue, or escalate the case without parsing generated prose.
The model card describes the design as “A typed decision model. One forward pass, no decoding loop, no parser, no retry.” That approach targets bounded decisions such as classification and routing rather than open-ended writing, conversation, or document creation.
Recommended Free Tools
Examples of decisions
- Policy checks and risk-band assignment
- Shell-command safety classification
- Rate-limit selection
- Refund-window decisions
- Queue or workflow choice
Developers can define the valid options for each decision and use the confidence score to send uncertain cases to a human reviewer or a larger model. The result is a predictable interface: downstream code receives one of the declared values instead of handling unexpected text.
#1 Best Overall
- POWERFUL COMPUTING: Advanced single board computer featuring high-speed LPDDR5 memory for superior processing capabilities and edge AI computing performance
- CONNECTIVITY: Multiple USB ports, HDMI output, and Ethernet connectivity provide versatile interface options for various applications
- COMPACT DESIGN: Space-efficient circuit board layout integrates powerful computing components in a single compact form factor
- DEVELOPMENT READY: Ideal platform for edge AI development, programming, and prototyping with comprehensive hardware interfaces
- EXPANDABILITY: Features multiple GPIO pins and standard connectors enabling extensive hardware expansion possibilities
Why edge execution matters for privacy
FLock says the checkpoint can run on laptops, smartphones, and other edge hardware, allowing an application to evaluate sensitive inputs locally rather than sending them to a hosted inference service. Keeping data on-device can reduce exposure during routine decisions, although it does not remove the need to secure the device, model files, logs, or application around it.
FLock’s broader partnership with Qwen combines Qwen models with federated learning, privacy-preserving computation, and distributed-ledger methods. The partnership announcement discusses potential applications in healthcare, finance, and education, where organizations may want to improve models without pooling raw records in one central system. Those partnership techniques are separate from the basic local inference workflow: running THIS/THAT locally does not by itself mean that an application is using federated learning or on-chain execution.
Rank #2
- [High performance] Quad-core ARM SoC up to 1. 8GHz with 3GB RAM- The Tinker Edge R features the Rockchip RK3399Pro SoC and Mali - T764 GPU along with 2GB of Dual Channel LPDDR4 memory for system, 1 GB LPDDR3 memory for NPU and 16GB eMMC flash
- [Gigabit Class networking]Tinker Edge R features a high speed GB LAN port for true Gigabit Class networking throughput along with 3x USB3.2 Gen1 Type-A. It also features onboard Wi-Fi & Bluetooth for robust IoT & Network connectivity
- [Open-source]The board will come with fully open-source kernel and support for multiple APIs, including OpenGL, Vulkan, OpenCL, OpenVX, TensorFlow Lite, Android NN, and Caffe
- [HD Audio & UHD video support] It supports 192/24bit HD Audio playback with automatic Audio jack detection as well as accelerated HD & UHD ( 4K ) video playback and supports HDMI CEC for seamless power on & off configurations
- [WiKi]For more information please refer to the product description, any technical issues after purchase please contact with our tech-support team: click "WayPonDEV" and ask a question. Package Content: 1x Tinker Edge R (3GB+16G eMMC); 2x Wi-FiVBT antenna cable; 1x Stand offset(4xScrew+4xHex); 2x Camera MIPI Convert cable (22P to 15P); 1 x Shielding bag; 1 x Quick start guide
Speed, size and published performance
| Specification or result | What FLock reports | How to interpret it |
|---|---|---|
| Parameter count | 1.88 billion | Small enough for the edge-oriented deployment FLock describes; actual memory needs depend on precision and runtime. |
| Architecture | Qwen3.5-style hybrid architecture | It uses Qwen3.5-2B as its foundation, with the decision-model behavior exposed through typed choices. |
| Generated output tokens | Zero | The model scores declared options rather than decoding a prose response. |
| Inference latency | About 31 milliseconds on a consumer GPU | A vendor-listed measurement; hardware, batching, precision and software settings will change the result. |
| Version 1.2 decision success | 87.8% across 1,710 complex decision tasks | A company-reported test result awaiting independent replication. |
| Optimization comparison | Up to 5× performance versus named larger models on specific tests | Not a general head-to-head result across all workloads; the tested tasks and conditions matter. |
The 31-millisecond figure and the reported accuracy and optimization gains should therefore be treated as published vendor results, not universal guarantees. Teams should measure their own schemas, hardware, languages, and error costs before replacing an existing model.
THIS/THAT versus a general-purpose cloud LLM
| Decision factor | THIS/THAT Model | General-purpose cloud LLM |
|---|---|---|
| Latency | Designed for a single local forward pass; FLock lists about 31 ms on a consumer GPU. | Depends on network travel, provider load, model size and generation length. |
| Output control | One declared option, index, probabilities and confidence. | Flexible text that may require schemas, parsing, retries or validation. |
| Task breadth | Bounded classifications, routing and policy decisions. | Open-ended generation, reasoning, summarization and many other tasks. |
| Data location | Can remain on a laptop, phone or edge device when run locally. | Normally processed in the provider’s environment when using a hosted endpoint. |
| Escalation | Confidence can trigger a human or larger-model review. | Requires the application to design and enforce its own confidence or review policy. |
| Cost model | Local inference avoids per-request cloud charges but uses the organization’s hardware and operations budget. | Hosted use typically follows provider pricing and usage; exact THIS/THAT API rates are not stated here. |
The sources do not establish independent, across-the-board head-to-head results for every row. The practical choice is workload-specific: use a typed edge model when the answer space is known and privacy or predictable latency dominates; use a general model when the task requires unrestricted language generation or broader reasoning.
Rank #3
- Supports access to online large model platforms and includes Edge Impulse object detection demo for real-time multi-object recognition
- Equipped with Xtensa dual-core LX7 processor (up to 240MHz), 8MB PSRAM, 16MB Flash, and dual-mode WF + BT LE
- Dual-microphone array with noise reduction and echo cancellation for high-quality voice processing
- Integrated audio input and output module, supporting AI speech interaction and voice recognition applications
- Onboard camera interface (DVP) and SPI / QSPI display interface for image capture, recognition, and external display connection
How developers can deploy it
Run the checkpoint locally
- Install the required Python environment and the Transformers stack described in the official Hugging Face model card.
- Load the THIS/THAT checkpoint through Transformers, then provide the state, question and permitted options expected by the model’s typed interface.
- Read the selected option, index, probabilities and confidence value from the forward-pass result.
- Define an escalation threshold before production. Route low-confidence outcomes to a human or a larger model instead of treating every prediction as an automatic approval.
- Benchmark on representative local hardware, including cold-start time, sustained throughput, memory use and behavior under your decision schemas.
Use hosted inference
FLock’s separate API Platform advertises an OpenAI-compatible interface, one API key, multiple modalities, usage-based billing and enterprise compliance options. A hosted endpoint can simplify operations when a team cannot manage local hardware, but the request then leaves the device. Confirm retention, regional processing, access controls and compliance terms for the specific account before sending regulated data.
Operational safeguards and limitations
- Declare options carefully: Mutually exclusive, complete choices reduce ambiguous routing. Include an explicit review or unknown option where the business process needs one.
- Calibrate confidence: A confidence score is useful for escalation only after testing how it relates to real error rates in your data.
- Protect local assets: On-device processing does not protect data exposed through telemetry, crash reports, application logs, backups or a compromised endpoint.
- Monitor drift: Policy changes, new attack patterns and changing customer behavior can make a previously reliable classifier stale.
- Keep a human path: High-impact financial, medical, employment or safety decisions should have review and appeal mechanisms appropriate to the risk.
- Validate the claims: Reproduce latency and accuracy on your own hardware and cases; the published 87.8% and 5× figures are vendor-reported.
What the launch means
FLock is positioning THIS/THAT as a specialized alternative to sending every small enterprise decision through a large cloud language model. Its combination of a 1.88-billion-parameter footprint, zero-token option scoring and local execution could fit applications that need fast, repeatable choices while limiting data movement. The trade-off is deliberate narrowness: teams must define the answer space, test calibration and build escalation logic, while open-ended work still belongs with a general-purpose model.
Quick Recap
Rank #4
- 30-in-1 No-Solder Sensor Board, Plug and Play: Integrates 30 functional sensors including temperature & humidity, ultrasonic ranging, gas and motion sensors. Innovative common board design requires no soldering or complex wiring, and comes with a full set of accessories like 128G SD card, adapter board and acrylic mounting plates for zero-threshold experiments
- 8MP Gimbal Camera & Dual Servos for Professional Visual AI: The Starter Kit is equipped with an IMX219 8MP monocular camera and a dual-servo gimbal, supporting face and target tracking, and is ideal for AI edge computing scenarios such as intelligent monitoring, robot navigation, and automated recognition
- 38 Step-by-Step Python Tutorials, From Beginner to Practical Application: The Jetson Orin Nano Starter Kit comes with 38 well-designed Python tutorials progressing from basic programming to vision practice, covering all key knowledge of sensor control, embedded development and AI visual recognition for both beginners and advanced learners
- 11.6-inch IPS HD Screen & AI Voice Interaction System: Built-in 1366*768 resolution IPS screen eliminates the need for an external monitor, enabling one-device experimentation and visual feedback. The exclusive AI voice interaction system supports intelligent Q&A and voice command control for natural human-computer dialogue
- Rich Expansion Interfaces & Portable All-in-One Design: Features 2x I2C, 1x UART and 2 IO expansion interfaces to meet personalized experiment expansion needs; a custom carrying case integrates all components (11.81×7.87×3.94 inch), allowing AI experiments and demonstrations anytime and anywhere
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
