Models added to LLMTR in October 2026: facts and guide

Short answer: eight new models, one API

LLMTR (llmtr.com) is a gateway that gives access to language models hosted in Türkiye and at global providers through one OpenAI-compatible API. The models added to its catalog in October 2026 include Glimpse (elorian/glimpse), Muse Spark Open (meta/muse-spark-open), Nemotron 3.5 Super VL (nvidia/nemotron-3.5-super-vl and the free nvidia/nemotron-3.5-super-vl-free), Moondream 3.1 (moondream/moondream3.1-9b-a2b), EmbeddingMagibu 200M (llmtr/embeddingmagibu-200m), Jev 1.14 (typesafe/jev-1.14.0) and Beam 1.0 (reflection/beam-1.0-501b-a23b). Liquid AI's decision model d1-ultra (liquid/d1-ultra) was added in October 2026 too. All of them are called at https://llmtr.com/v1: chat models use the /v1/chat/completions endpoint, the embedding model /v1/embeddings and the decision models /v1/systemone.

Other models were added in October as well; the full, current list is in the model catalog. The free Qwen3.8 27B row in the table (qwen/qwen3.8-27b-free) was added before October and is listed here as a free row you can use alongside the vision models added in October.

Which models did LLMTR add in October 2026?

The table gives each row's id, endpoint, input types, context window and price per 1M tokens as the catalog lists them. Dated prices show both periods; each request is billed at the catalog price in effect when it is processed. The capabilities below are taken from the models' LLMTR cards.

Models that understand images and video

Glimpse is Elorian's visual reasoning agent. According to its card, it can run code and search the web before it replies, so an answer can take seconds to minutes; send stream: true for longer tasks. The agent's steps are billed as output tokens. The text, images and videos you send may be retained by the provider and used to train Elorian's models; do not send sensitive, personal or customer data.

Muse Spark Open is the mixture-of-experts (MoE) reasoning model Meta released with open weights: 816 billion parameters in total, about 42 billion active per token. It is trained for agentic coding and returns tool calls and JSON Schema output. It reasons before every answer; reasoning_effort selects low, medium, high or xhigh, and without the field the model reasons at a high effort.

Nemotron 3.5 Super VL is NVIDIA's open multimodal MoE model (120 billion total, 12 billion active parameters). It offers tool calling and JSON Schema output; reasoning is on by default and can be switched off per request. On both rows prompts may be logged by the provider; the text, images and videos sent to the free row may also be used to improve the model, so do not send sensitive, personal or customer data there.

Moondream 3.1 is Moondream's small vision-language model with 9 billion parameters (2 billion active per token), used for visual question answering, captions, reading text in images and visual reasoning. Reasoning is off by default and turns on with reasoning: true. The card of the free Qwen3.8 27B row lists image input, tool calling and JSON output; it produces at most 16,384 tokens per response. On this row, prompts and completions may be logged by the provider and used to improve the model.

• Elorian, Introducing Glimpse · 2026-10-17
• Meta, Build with Muse Spark Open · 2026-10-17
• Moondream, Moondream 3.1 model sayfası · 2026-10-17

Embedding and decision models

EmbeddingMagibu 200M is an embedding model trained for Turkish text and hosted in Turkey; it turns text into a normalized 768-dimensional vector. Documents are sent with prompt_name set to document and search queries with query. During the free period that runs until 20 October 2026, 23:59 (Turkey time), each account has a daily allowance of 500,000 input tokens.

d1-ultra is Liquid AI's decision model: it evaluates typed questions against a state you send as text or JSON, with optional images, and returns for each question a choice, a score or the probability that a statement is true; it generates no free text. Every question is read and billed along with the whole state and all images. The text and images you send may be retained by the provider and used to train Liquid's models; do not send sensitive, personal or customer data.

Jev 1.14 is the release of TypeSafe's decision model Jev that became available on 14 October 2026. It returns a choice, a score or a probability for typed questions, and with a Box question it finds where an object is in an image. Images and the long-context mode are experimental features the provider can still change. The previous release, Jev 1.13, stays on its own model id.

• Hugging Face, alibayram/embeddingmagibu-200m model kartı · 2026-10-17
• Liquid AI, d1-ultra duyurusu · 2026-10-17
• TypeSafe, Models · 2026-10-17

Beam 1.0: free until 22 October 2026, 23:59, then retired

Beam 1.0 is Reflection AI's mixture-of-experts model for coding, reasoning and agentic workloads, with 501 billion total and 23 billion active parameters. Reflection says this early version is open to a select group of users. On LLMTR it accepts text only, and tool calling and structured output with JSON Schema are supported. The model reasons before every answer and reasoning cannot be turned off; reasoning_effort can be low, medium, high, xhigh or max, and max applies when it is omitted.

Every price rule is $0: the model is served free with a daily quota until 22 October 2026, 23:59 (Turkey time). From 23 October 2026, 00:00 (Turkey time) it is retired; requests for this id return HTTP 410 model_retired, and the error message says the free period has ended and that LLMTR has no other Reflection model. Prompts and completions may be logged by the provider and used to improve the model; do not send sensitive, personal or customer data.

• Reflection AI, Introducing Beam · 2026-10-17
• Reflection AI, Models · 2026-10-17

Which one for which job?

Video: Glimpse takes a video as an https URL and Nemotron 3.5 Super VL as a base64 data URL; Muse Spark Open accepts audio as well as video. For images only, Moondream 3.1 and the free Qwen3.8 27B row can be used too.

Long context: Muse Spark Open offers a 1,048,576-token context and the free Nemotron row 1,000,000 tokens; Beam 1.0's 1,048,576-token context is available until 22 October 2026, 23:59 (Turkey time). For Turkish search and RAG, EmbeddingMagibu 200M is on /v1/embeddings. When routing or classification needs a probability instead of free text, look at the decision models on /v1/systemone; to locate an object in an image, use Jev 1.14's Box question. The code below sends a chat and an embedding request with one client.

One OpenAI client: a chat request to meta/muse-spark-open and an embedding request to llmtr/embeddingmagibu-200m.

import os

from openai import OpenAI

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

# Chat models use /v1/chat/completions.
reply = client.chat.completions.create(
    model="meta/muse-spark-open",
    messages=[{"role": "user", "content": "List three checks for a pull request that adds a database migration."}],
    reasoning_effort="medium",
    max_tokens=4096,
)
print(reply.choices[0].message.content)

# The embedding model uses /v1/embeddings; prompt_name goes in extra_body.
vectors = client.embeddings.create(
    model="llmtr/embeddingmagibu-200m",
    input=["Faturalar her ayın ilk iş günü kesilir."],
    extra_body={"prompt_name": "document"},
)
print(len(vectors.data[0].embedding))

Free rows and getting started

On the free Nemotron 3.5 Super VL row, the free Qwen3.8 27B row and Beam 1.0 until 22 October 2026, 23:59 (Turkey time), HTTP 429 means one of the free usage limits has been reached; the limit can be daily or per minute, and the response carries a Retry-After header.

To get started, create an account at llmtr.com, top up your balance under Dashboard > Billing for paid models, create a key under Dashboard > API Keys, store it in the LLMTR_API_KEY environment variable and set base_url to https://llmtr.com/v1 in your OpenAI-compatible SDK. The same key is valid for every model in the catalog, and the other articles in this series cover each model separately.

Call a model added in October 2026 through LLMTR

Create an account, create an API key and send the first request through https://llmtr.com/v1 to the endpoint that matches the model type.

1. Account. Create an account at llmtr.com; for paid models, top up your balance under Dashboard > Billing.
2. API key. Create a key under Dashboard > API Keys. The raw key is shown once; store it as the LLMTR_API_KEY environment variable.
3. First request. Set base_url to https://llmtr.com/v1; send chat models to /v1/chat/completions, EmbeddingMagibu 200M to /v1/embeddings and decision models to /v1/systemone, with the model id.

Frequently asked questions

Which of the models LLMTR added in October 2026 are free?

nvidia/nemotron-3.5-super-vl-free is free with a daily quota. Beam 1.0 is free until 22 October 2026, 23:59 (Turkey time) and retired from 23 October 2026, 00:00 (Turkey time). EmbeddingMagibu 200M costs $0 from 15 October 2026, 00:00 to 20 October 2026, 23:59 (Turkey time) and $0.01 per 1M input tokens from 21 October 2026, 00:00 (Turkey time). qwen/qwen3.8-27b-free, added before October, is also free with a daily quota.

Can I use Beam 1.0 after 23 October 2026?

No. From 23 October 2026, 00:00 (Turkey time), requests for reflection/beam-1.0-501b-a23b return HTTP 410 model_retired; the error message says LLMTR has no other Reflection model. Choose an active model from the catalog.

Are all of these models called with the same API key?

Yes. All of them are called at https://llmtr.com/v1 with the same LLMTR key; the endpoint depends on the model type: chat models use /v1/chat/completions, EmbeddingMagibu 200M /v1/embeddings and the decision models /v1/systemone.

Related posts

• Glimpse API: a reasoning agent for images and video
• Video analysis API: asking Glimpse about a video
• Muse Spark Open API: OpenAI SDK usage and pricing
• Send images, video and audio to Muse Spark Open
• Free Nemotron 3.5 Super VL API: image and video input
• Nemotron 3.5 Super VL: free vs paid row
• Send images to the free Qwen3.8 27B: API guide
• Moondream 3.1 API: base64 images, reasoning and pricing
• Turkish embedding API: a guide to EmbeddingMagibu 200M
• Turkish semantic search with EmbeddingMagibu, step by step
• Liquid d1-ultra API: probability decisions on /v1/systemone
• Jev 1.14 Box: finding an object's location in an image
• Glimpse reasoning_effort and cost: low, medium, xhigh
• Muse Spark Open reasoning_effort: levels and cost
• Tool-calling agent on the free Nemotron 3.5 Super VL
• JSON output from the free Qwen3.8 27B: json_object guide
• Moondream 3.1 or Moondream 3 Preview? Differences and price
• Categorizing a product catalog with d1-ultra, with review
• Jev 1.14 or Jev 1.13? Differences, pricing and migration

• Model catalog
• Chat Completions guide
• Embeddings documentation
• Structured Decisions guide
• Pricing