Model comparison · 2026-10-17

Arabic models on LLMTR: Jais 3, Jais ASR and K2 Horizon

The IFM models added to LLMTR: Jais 3 for Arabic and English, Jais ASR for Arabic speech-to-text, and K2 Horizon. Which suits which job, allowances and dates.

Short answer: three IFM models, one API

LLMTR is a gateway that gives access to language models from different providers through one OpenAI-compatible API. In October 2026 three models from the Institute of Foundation Models (IFM) at MBZUAI were added to the catalog. Jais 3 is the first Arabic-focused language model added to the LLMTR catalog; next to it came Jais ASR, which transcribes Arabic recordings, and IFM's general model K2 Horizon.

IFM and Inception42 announced Jais 3 on 15 October 2026. According to the announcement, Jais 3 is built on IFM's K2 Horizon model family, and the release paired six open Jais 3 language models between 900 million and 375 billion parameters with a speech recognition model. The three LLMTR rows come from that release.

Which Arabic models are on LLMTR?

The table shows the three rows, their endpoints and their daily allowances. The two chat models are called on /v1/chat/completions and Jais ASR on /v1/audio/transcriptions. The same API key works for all three.

To see the IFM rows of the catalog in code, read the /v1/models list. The Python example below fetches the list with the OpenAI SDK and prints the ids that start with ifm/. Each row's details are on its LLMTR model page and in the IFM Jais documentation.

IFM models on LLMTR, as of 17 October 2026
Model id	Job	Endpoint	Daily allowance (smaller on a new account's first day)
ifm/jais-3-mova-36b-a4b	Arabic and English chat, tool calling, JSON	/v1/chat/completions	200,000 tokens
ifm/jais-asr	Transcribing Arabic recordings, dialect label	/v1/audio/transcriptions	60 minutes of audio
ifm/k2-horizon-375b-a23b	General model that reasons before every answer, tool calling, JSON	/v1/chat/completions	200,000 tokens

An example that fetches LLMTR's /v1/models list with the OpenAI Python SDK and prints the model ids that start with ifm/.

import os

from openai import OpenAI

client = OpenAI(base_url="https://llmtr.com/v1", api_key=os.environ["LLMTR_API_KEY"])

for model in client.models.list():
    if model.id.startswith("ifm/"):
        print(model.id)

Which model for which job

If you work with Arabic text, for example answering, classifying or turning Arabic customer messages into JSON, start with Jais 3. IFM says the model is designed for Arabic and English assistant and agentic applications; on LLMTR tool calling works with tool_choice auto, and reasoning is chosen between high and low.

If you have recordings, Jais ASR gives the transcript and the recording's language or dialect label; Modern Standard Arabic and the dialects of 19 countries are recognised. In a call-centre flow, transcribing the recording with Jais ASR and having Jais 3 summarise the transcript combines both models in one flow.

If your agent framework sends forced tool calls, or you have English jobs that need multi-step reasoning, use K2 Horizon. The model reasons before every answer; reasoning_effort selects medium or high, and max_tokens should leave room for the reasoning.

The free period and after

All three rows are free until 5 November 2026, 23:59 (TRT). During this period every account has 200,000 tokens a day each for Jais 3 and K2 Horizon, and 60 minutes of audio a day for Jais ASR. The allowances renew at 00:00 (TRT), and new accounts start with a smaller allowance on their first day. When an allowance is used up the API returns 429; the Retry-After header tells you when you can try again.

From 6 November 2026, 00:00 all three models return 410 model_retired. Requests are not silently routed to another model, and nothing is charged to your account until you change the model id yourself. The model pages stay in the catalog.

According to the data note on the LLMTR model pages, prompts, recordings and responses may be logged by the provider and used to improve the model; do not send sensitive, personal or customer data.

The other articles in the series

For Jais 3: a first Arabic request, an Arabic agent that calls tools, JSON output from Arabic text and the difference between reasoning_effort low and high. For K2 Horizon: using the API, and setting reasoning_effort and max_tokens. To see the two models side by side, the Jais 3 and K2 Horizon comparison. For Jais ASR: transcribing Arabic recordings and dialect identification.

Each article carries its own code example and a table of the values on LLMTR. The full list of request fields and the error codes is on the IFM Jais documentation page.

Get started with the IFM models on LLMTR

Create an API key, pick the IFM model that fits your job and send a first request to its endpoint.

API key. Create an account at llmtr.com, create a key under Dashboard > API Keys and keep it in the LLMTR_API_KEY environment variable.
Pick a model. Choose ifm/jais-3-mova-36b-a4b for Arabic text, ifm/jais-asr for recordings and ifm/k2-horizon-375b-a23b for forced tool calls.
First request. Call the chat models at https://llmtr.com/v1/chat/completions and Jais ASR at https://llmtr.com/v1/audio/transcriptions.

Frequently asked questions

Is there an Arabic language model on LLMTR?

Yes. ifm/jais-3-mova-36b-a4b is the Jais 3 model built for Arabic and English, and it is the first Arabic-focused language model added to the LLMTR catalog. For Arabic recordings there is also ifm/jais-asr.

How long are the IFM models free?

Until 5 November 2026, 23:59 (TRT). From 6 November 2026, 00:00 all three models return 410 model_retired, and requests are not routed to another model.

Do the three models work with the same API key?

Yes. The key you create on LLMTR works for all three. The chat models are called on /v1/chat/completions and Jais ASR on /v1/audio/transcriptions.