Products

What we ship.

Everything we release is open-weight, with the tokenizer, the evaluation and the write-up attached. One product so far — the next ones will be listed here as they land.

fihrIA-ASR0.6B & 1.7B

Speech recognition · 2026

0.6B released

Open-weight encoder–decoder speech recognition for Moroccan Darija. The 0.6B model is on Hugging Face now; the 1.7B follows.

  • 0.6BReleasedQ4_K_M and BF16 GGUF · 721 MB at 4-bit
  • 1.7BComing soonWeights, GGUF and release notes in preparation

In the workshop

More speech and language models for Morocco are in progress. Releases are announced on the blog first.

Follow the blog