The $15 chip now runs a language model
the $15 chip now runs a language model: a developer got a 28.9-million-parameter model running offline on an esp32-s3 at 9.5 tokens a second, and microsoft’s phi-4 edge hits 92% of gpt-4o’s score on a laptop chip.
the frontier stopped being about the biggest model and started being about who runs without a network.
Context
The slvDev/esp32-ai README describes a 28.9 million parameter language model generating text on an ESP32-S3, a microcontroller costing about $8, with nothing sent to a server. It lists about 9.5 tokens per second end to end, 14.9MB at 4-bit, and 25M of the parameters in a flash lookup table.
The README says it uses Per-Layer Embeddings from Google's Gemma models and that the last model on a chip like this had 260 thousand parameters. A raw README variant states 9.88 tokens per second, so the exact figure differs between versions.
The 28.9 million parameters, the ESP32-S3 and about 9.5 tokens per second match the README. The post says $15; the README says about $8. The claim that Microsoft's phi-4 edge hits 92 percent of GPT-4o's score on a laptop chip was not found; the sources read are the phi-4 technical report (Dec 12, 2024) for a 14-billion parameter model, so the 92 percent figure is not seen in the sources read, so unsupported here, not refuted.
Related work
- GitHub: slvDev/esp32-ai README ↗Project page.
- DeepWiki: esp32-ai inference profiling ↗Jul 26, 2026.
- Microsoft Research: Phi-4 Technical Report ↗Dec 12, 2024.
Watch next
- Find the source of the phi-4 edge 92 percent claim.
Sources
- slvDev, esp32-ai READMEgithub.com
- DeepWiki, esp32-ai profilingdeepwiki.com
- Microsoft Research, Phi-4 Technical Report, Dec 12, 2024microsoft.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 2 October 2026 at 02:16 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →