One correction up front: ESP32-S3 does not support Wi-Fi 6. S3 is Wi-Fi 4 (2.4GHz) plus BLE 5.0; Wi-Fi 6 together with Thread/Zigbee/Matter belongs to ESP32-C6; coin-cell class low power is ESP32-H2. Get this wrong in a spec and a competent reviewer will catch it on page one.
1. Where the four parts actually sit
| Part | Radio | Thread / Zigbee / Matter | Compute | Key peripherals | Typical product | Avoid it for |
|---|---|---|---|---|---|---|
| ESP32-S3 | Wi-Fi 4 (2.4GHz) + BLE 5.0 | No | Xtensa dual-core 240MHz + vector extensions | USB OTG, DVP camera, parallel LCD, OPI PSRAM | Offline voice, AI toys, small panels, doorbells | No Wi-Fi 6, no Thread/Zigbee |
| ESP32-C6 | Wi-Fi 6 (2.4GHz) + BLE 5.3 | Thread / Zigbee / Matter | RISC-V 32-bit 160MHz | 802.15.4 coexistence | Matter gateways, export smart-home nodes | Less compute and no camera path vs S3 |
| ESP32-H2 | No Wi-Fi | Thread / Zigbee | RISC-V 96MHz | Ultra-low sleep current | Coin-cell sensors, locks, switches | Cannot reach the cloud directly |
| ESP32-P4 | Radio usually external | Depends on design | RISC-V dual-core 400MHz + security | Rich high-speed IO and display | High-end HMI and control panels | Highest cost and complexity |
Disclaimer: specifications are compiled from public datasheets and third-party benchmarks; always validate against the vendor datasheet and real board testing. Prices and price movements reflect publicly visible channels — the quotation and contract govern (some figures remain TBC).
2. Three scenarios, three answers
Offline voice and AI toys
ESP32-S3. Wake-word runs locally, and PSRAM plus larger Flash hold acoustic models and audio assets. Our own Circular AI Voice Interaction Core Board (ESP32-S3) and the AI Speaking Practice Device sit on exactly this path.
Video doorbells and small appliances with screens
Still S3: DVP camera interface, parallel LCD and USB OTG all in one part. Watch the PSRAM option, frame rate, and 2.4GHz congestion behaviour.
Matter gateways and export smart-home lines
ESP32-C6. Coexisting Wi-Fi 6 and 802.15.4 let it serve as Thread border router and Wi-Fi node simultaneously, which S3 cannot do structurally.
3. Three things teams under-estimate before production
- Certification path: pick a module that already carries radio approval instead of discovering the RF bill at device level.
- PSRAM and Flash options: voice and GUI workloads are PSRAM sensitive — fix the capacity during selection.
- Battery life: measure sleep current and the wake path together; do not judge from a demo unit.
4. The rule we use
IoT projects rarely fail on chip performance. They fail because the radio does not match the scenario. Split the brief into three questions — camera or not, screen or not, Thread/Matter or not — and the part number usually settles itself.
FAQ
- Q：Which Wi-Fi does ESP32-S3 actually support?
- Wi-Fi 4 (802.11 b/g/n, 2.4GHz single band) at up to 150Mbps theoretical, plus Bluetooth LE 5.0. Wi-Fi 6 (802.11ax) belongs to ESP32-C6. Check the datasheet before this lands in a specification — it is the first thing a knowledgeable buyer notices.
- Q：What should I use for offline wake-word products?
- ESP32-S3. Vector extensions, the ULP coprocessor and USB OTG suit local wake-word and light voice interaction at low cost. If the product also needs a camera and a small screen, such as a video doorbell, S3 remains the smoothest path.
- Q：We need Matter for export markets. Now what?
- Use ESP32-C6. Combining Wi-Fi 6 with 802.15.4 lets it act as a Thread border router and a Wi-Fi node at once, which S3 structurally cannot do.
- Q：Is the module already certified?
- Prefer a module version that already carries SRRC / CE / FCC approval so you avoid re-certifying the radio, but the finished device still needs market-specific EMC, safety and SAR assessment. Confirm the exact certificate scope for the module you buy.