IoT Smart Modules in 2026: When the Modem Becomes the Computer
A cellular smart module runs a full application processor alongside the radio, so one component can be both the computer and the mobile connection. Here are four current examples and where each one fits.
A smart module is a cellular module that also runs a full application processor under Linux or Android, so a single part provides both the computer and the mobile connection. This guide compares four current smart modules, from an entry-tier LTE Cat 4 part to a 48 TOPS edge-AI flagship, and shows where each one earns its place. Sourcing help sits under every module.
What a smart module actually is
For years the cellular module had one job. Your processor ran the application, and the module connected it to the mobile network. A smart module blurs that line. It pairs the cellular radio with a proper application processor, memory, storage and an operating system, so the module itself can run the software the product needs.
This has become practical because the silicon underneath has moved up the computing stack. The parts below are built on Qualcomm IoT platforms that carry multi-core 64-bit CPUs, capable GPUs and, at the upper end, dedicated neural processing for on-device AI. In other words, several of these modules are closer to a small embedded computer with a modem inside than to a modem waiting for instructions.
That shift is why product designers are starting to ask whether a separate single-board computer, or a separate 4G router, still needs to sit next to the module at all. The full architecture argument is covered in smart module vs SBC vs router. This page is the reference layer beneath it: what the modules themselves are, and how they compare.
A plain cellular module is a modem. A smart module is a modem plus an application processor and operating system. The difference matters most when you are deciding how many separate boards a connected product actually needs.
Four smart modules available in 2026
These four span the range, from an entry multimedia module to a flagship edge-AI part. All specifications below are taken from the manufacturers' own product pages, listed in the sources at the foot of the article.

Telit Cinterion SE250B4
A second-generation Android and Linux smart module built for touch-screen, camera-equipped devices that need modest local compute and reliable Cat 4 connectivity in one package.
- SoCQualcomm QCM2290
- CPU / GPUQuad-core Arm Cortex-A53 at 2.0 GHz, Adreno 702 GPU
- CellularLTE Cat 4 (150 Mbps down / 50 Mbps up)
- WirelessWi-Fi 802.11a/b/g/n/ac, Bluetooth LE 5.0, multi-constellation GNSS
- Edge AINo dedicated NPU
- OSAndroid and Linux
- Form factorLGA, 41 x 43 mm
Best for: mobile point-of-sale, smart cash registers, vending machines, alarm panels, handheld terminals and surveillance where multimedia matters more than heavy AI.

Quectel SC696S
A multi-mode LTE Cat 4 smart module with built-in Linux, aimed at industrial and consumer devices that need high data rates, multiple cameras and a rich interface set in a single part.
- SoCQualcomm QCM6125
- CPU / GPU64-bit octa-core Arm Kryo 260, Adreno 610 GPU
- CellularLTE Cat 4, UMTS/HSPA+, GSM/GPRS/EDGE, MIMO
- WirelessMulti-constellation GNSS (GPS, GLONASS, BeiDou, Galileo, QZSS, SBAS); Wi-Fi/Bluetooth on the -WF variant
- Edge AINo dedicated NPU
- MultimediaUp to 6 cameras (2 concurrent), 4K video at 30 fps, MIPI DSI display
- OSLinux
Best for: smart POS, industrial handhelds, audio and video recorders, smart hardware and, in Quectel's own words, vending machines and logistics cabinets.

Fibocom SC171
A 5G Sub-6 smart module on the Qualcomm QCM6490 platform, built for higher-end AIoT devices that need edge inference, multi-camera input and 5G connectivity together.
- SoCQualcomm QCM6490
- CPU / GPUOcta-core (1 x 2.7 GHz + 3 x 2.4 GHz + 4 x 1.95 GHz), Adreno 642L GPU
- Cellular5G NR Sub-6 SA/NSA, backward compatible with 4G/3G; DL 4×4 / UL 2×2 MIMO
- WirelessWi-Fi 6/6E, Bluetooth 5.2, multi-constellation GNSS
- Edge AIUp to 12 TOPS
- OSAndroid
Best for: 5G AIoT, industrial handhelds, body-worn and security cameras, mobile DVRs and robotics where edge AI and 5G are both needed.

Quectel SG882G-AP
A flagship Android and Linux smart module on Qualcomm's QCS8550 IoT chipset. This is a compute-led part: cellular is added via a companion modem, while the module supplies serious on-device AI and multimedia.
- SoCQualcomm QCS8550 (flagship IoT chipset)
- CPU / GPUOcta-core Kryo CPU, Adreno 740 GPU, plus DPU, VPU, Hexagon DSP and Spectra ISP
- CellularAdded via companion modem (no integrated cellular)
- Edge AIUp to 48 TOPS NPU
- Memory12 GB LPDDR5X + 256 GB UFS
- OSAndroid 13 / Linux (kernel 5.15)
- Form factor33.0 x 39.0 x 3.85 mm
Best for: vision-heavy edge AI, intelligent retail, smart safety, robotics, AR/VR and video systems where raw local compute leads the design and cellular is a companion.
Smart module comparison
The same four modules side by side. Read across a row for a single module's profile, or down a column to compare one attribute across the range.
| Module | SoC | Cellular | Peak edge AI | OS | Typical use |
|---|---|---|---|---|---|
| Telit Cinterion SE250B4 | Qualcomm QCM2290 (quad Cortex-A53) | LTE Cat 4 | No dedicated NPU | Android / Linux | mPOS, cash registers, vending |
| Quectel SC696S | Qualcomm QCM6125 (octa Kryo 260) | LTE Cat 4 | No dedicated NPU | Linux | POS, handhelds, vending, recorders |
| Fibocom SC171 | Qualcomm QCM6490 (octa-core) | 5G Sub-6 SA/NSA + 4G/3G | Up to 12 TOPS | Android | 5G AIoT, cameras, robotics |
| Quectel SG882G-AP | Qualcomm QCS8550 (octa Kryo) | Via companion modem | Up to 48 TOPS | Android 13 / Linux | Edge AI, vision, intelligent retail |
Peak TOPS is a first-order signal, not a promise. Real inference throughput depends on the workload and how well the Qualcomm AI stack and your application use the NPU. Treat the number as a shortlist filter, then test on your actual model.
How to choose between them
The tiers map fairly cleanly to intent. If the device is a payment terminal, till or vending controller that needs a screen, a camera or two and steady Cat 4 data, the SE250B4 or SC696S cover it without paying for AI silicon you will not use. If the product needs 5G and on-device inference, such as a connected camera or a handheld doing recognition in the field, the SC171 is the natural step up. If the design is vision-led and compute is the point, the SG882G-AP behaves more like an edge-AI computer, with cellular added alongside rather than built in.
One decision sits above the module choice: how connectivity is provisioned. Embedding a module deep inside a sealed product raises the question of what happens to the SIM. That is where eSIM and the GSMA's IoT eSIM standard come in, letting the connectivity profile be provisioned and changed remotely rather than physically. The current state of that standard is covered in our SGP.32 mid-2026 state of play, and the architecture trade-off between a module, an SBC and a router is worked through in smart module vs SBC vs router.
Frequently asked questions
What is a smart module?
How is a smart module different from a cellular modem?
Do smart modules support eSIM and SGP.32?
Can a smart module replace an SBC and a 4G router?
Which smart module is best for edge AI?
Specifying modules for a build?
Tell us the deployment and we will help you shortlist the right smart module and connectivity approach for it.



