Skip to content
Back to blog list
Ornith-1.5-122B Tanıtıldı: MIT Lisanslı MoE Mimarili Model
AI Tools

Ornith-1.5-122B Introduced: MIT Licensed MoE Architecture Model

Automatically translated
Can Kepenekoğlu
Can KepenekoğluWriter
5 min read
The largest feature of Ornith-1.5-122B does not activate all parameters during each process.

Artificial intelligence continues the competition of open weight models. Ornith AI released the new model Ornith-1.5-122B. Model uses Mixture-of-Experts (MoE) architecture and is distributed with MIT license. 122 billion parameter model aims to combine high model capacity with efficient trading resources thanks to the mixture-of-Experts architecture. The model designed specifically for software development, complex problem solving, autonomous artificial intelligence agents attracts attention with MIT license.

Ornith-1.5-122B test

Ornith-1.5-122BThe featured feature of the prefix does not activate all parameters during each process. MoE architecture activates artificial intelligence agents according to the type of incoming request. Thus, it reduces calculation load. Although the model has a large scale structure, it uses the process resources controlled. Ornith AI does indicate that the system can fight against competitors in coding and agent-based tests. However, these results consist of their own measurements of the company and there is no independent verification yet. Independent tests will not be correct to say that the model is more successful than all competitors.

What is Ornith-1.5-122B Properties?

Ornith-1.5 family started three models in August 2026: Ornith-1.5-397B, Ornith-1.5-35B-A3B and Ornith-1.5-9B. 122B model fills the gap between 9B light model with 397B large model as the fourth member of the family. Developers positions the model as a reasoning system that can analyze complex tasks, write code, interact with other tools, except for chatting and creating text.

Total parameter count was announced in 122 billion. This capacity helps long software projects evaluate extensive document studies in a single study session.

The outstanding features of Ornith-1.5-122B are:

122 billion parameters: It offers large-scale model capacity.

  • MoE architecture: Every transaction uses selected specialist networks.

  • 262K context window: It can work on long texts and code bases.

  • Mental execution capabilities: Ability to create solutions steps by analyzing complex tasks.

  • Vehicle Use: It can interact with external software and tools.

  • MIT license: Allows outdoor weights to be used in the framework of license conditions.

  • Mixture of experts çalışma şeması

    What is MoE Architecture? How Does It Work in Artificial Intelligence?

    Mixture-of-Experts is a architecture that only works by a group of specialist networks according to input. In traditional intensive models, the large part of the system is actively involved in calculation. MoE architecture is composed of smaller networks that specialize in different tasks. A guiding mechanism determines which specialist network will work by analyzing the incoming request. For example, code writing can benefit from different experts for tasks such as mathematical problem solving or text analysis. The main advantage of this approach reduces the amount of calculation used per transaction while maintaining the total capacity of the model. Ornith-1.5-122B also uses this architecture.

    Home MoE architecture There is an important technical limit. The reduction of the number of active parameters does not eliminate the need to keep all weights in memory. So the amount of memory required for the operation of the system is changed according to the used level, method.

    Total number of parameters of Ornith-1.5-122B is 122 billion. Some directories write 125,1 billion. The difference depends probably on the number method, but this is an estimate. For the number of active parameters per token, different figures are getting.

    What Results did they take in coding tests?

    Ornith AI, software development and autonomous agent wounds from different test applications to measure performance. The results that the company has announced are:

    Home k

    Ornith-1.5-122B

    Qwen3.8-Flash-Nex

    TerminalBench 2. 1

    81,2

    83,4

    SWE-Ich Pro

    76,9

    43,9

    SWE-Ich Verified

    84,5

    82,8

    Home

    80,0

    78,0

    These results show that the model aims to provide balanced performance in different tasks. But the test environment can affect the results of tools, commands and evaluation method used.

    bilgisayar programcısı ve yapay zeka ajanı

    What Does MIT Licensed AI Models Provide to Developers?

    The publication of Ornith-1.5-122B with MIT license is one of the outstanding advantages. MIT offers developers a wide range of freedoms in terms of using, modifying, distributing and evaluation in commercial projects. That’s why MIT’s licensed models are among the attractive options that companies can integrate into products. Model weights Home Hugging Face Sharing via the developers allows developers to create a working environment on their own servers. In addition, OpenAI compatible API servers are installed with infrastructures such as vLLM and SGLang.

    On the other hand, licensing conditions of the components that are based on the model should also be examined. It is necessary to look at the licensing obligations of the basic model rather than just looking at the license specified in the card.

    Ornith-1.5-122B aims to show that a powerful MoE model can be operated with modest hardware. As the number of active parameters is kept low, the cost of withdrawal offers the advantage. Nevertheless, the correcting of the official benchmark results with independent tests will determine the actual success of the model.

    yapay zeka ajanlarının gelişimi

    Frequently Asked Questions about Ornith-1.5-122B

    1. What is Ornith-1.5-122B?

    The Ornith team publishes, Mixture-of-Experts (MoE) is an open-weight artificial intelligence model. It is distributed with MIT license and accessible in Hugging Face. The fourth model of Ornith-1.5 family starting in August 2026.

    2. 2. What is MoE architecture?

    A architecture that only works by a group of experts. While the total number of parameters is large, only one part of the model in each request is activated. This reduces calculation load and memory consumption.

    3. How many parameters are the model?

    Total 122 billion parameters are reported. Some model directories write 125,1 billion. For the number of active parameters per token, resources give different figures from about 10 billion to 12,7 billion.

    4. What areas do you stand out?

    Coding, agent tasks, vehicle use and JSON output. Difficult tasks may require more than 32 thousand output tokens while explaining the mode.

    5. What does MIT license mean?

    An unrestricted license that releases commercial use and replacement. Weights are accessible. However, according to a model directory, the educational code is marked as “notibility” and the model is finely adjusted. You need additional information for a full transparency assessment.

    6. What are known limitations?

    A news site in Japanese tests was transferred to the search of Chinese characters and language verification. It does not seem a collective knowledge yet about Turkish performance.

    Artificial intelligence continues the competition of open weight models. Ornith AI released the new model Ornith-1.5-122B. Model uses Mixture-of-Experts (MoE) architecture and is distributed with MIT license. 122 billion parameter model aims to combine high model capacity with efficient trading resources thanks to the mixture-of-Experts architecture. The model designed specifically for software development, complex problem solving, autonomous artificial intelligence agents attracts attention with MIT license. Ornith-1.5-122B The featured feature of the prefix does not activate all parameters during each process. MoE architecture activates artificial intelligence agents according to the type of incoming request. Thus, it reduces calculation load. Although the model has a large scale structure, it uses the process resources controlled. Ornith AI does indicate that the system can fight against competitors in coding and agent-based tests. However, these results consist of their own measurements of the company and there is no independent verification yet. Independent tests will not be correct to say that the model is more successful than all competitors. What is Ornith-1.5-122B Properties? Ornith-1.5 family started three models in August 2026: Ornith-1.5-397B, Ornith-1.5-35B-A3B and Ornith-1.5-9B. 122B model fills the gap between 9B light model with 397B large model as the fourth member of the family. Developers positions the model as a reasoning system that can analyze complex tasks, write code, interact with other tools, except for chatting and creating text. Total parameter count was announced in 122 billion. This capacity helps long software projects evaluate extensive document studies in a single study session. The outstanding features of Ornith-1.5-122B are: 122 billion parameters: It offers large-scale model capacity. MoE architecture: Every transaction uses selected specialist networks. 262K context window: It can work on long texts and code bases. Mental execution capabilities: Ability to create solutions steps by analyzing complex tasks. Vehicle Use: It can interact with external software and tools. MIT license: Allows outdoor weights to be used in the framework of license conditions. What is MoE Architecture? How Does It Work in Artificial Intelligence? Mixture-of-Experts is a architecture that only works by a group of specialist networks according to input. In traditional intensive models, the large part of the system is actively involved in calculation. MoE architecture is composed of smaller networks that specialize in different tasks. A guiding mechanism determines which specialist network will work by analyzing the incoming request. For example, code writing can benefit from different experts for tasks such as mathematical problem solving or text analysis. The main advantage of this approach reduces the amount of calculation used per transaction while maintaining the total capacity of the model. Ornith-1.5-122B also uses this architecture. Home MoE architecture There is an important technical limit. The reduction of the number of active parameters does not eliminate the need to keep all weights in memory. So the amount of memory required for the operation of the system is changed according to the used level, method. Total number of parameters of Ornith-1.5-122B is 122 billion. Some directories write 125,1 billion. The difference depends probably on the number method, but this is an estimate. For the number of active parameters per token, different figures are getting. What Results did they take in coding tests? Ornith AI, software development and autonomous agent wounds from different test applications to measure performance. The results that the company has announced are: Home k Ornith-1.5-122B Qwen3.8-Flash-Nex TerminalBench 2. 1 81,2 83,4 SWE-Ich Pro 76,9 43,9 SWE-Ich Verified 84,5 82,8 Home 80,0 78,0 These results show that the model aims to provide balanced performance in different tasks. But the test environment can affect the results of tools, commands and evaluation method used. What Does MIT Licensed AI Models Provide to Developers? The publication of Ornith-1.5-122B with MIT license is one of the outstanding advantages. MIT offers developers a wide range of freedoms in terms of using, modifying, distributing and evaluation in commercial projects. That’s why MIT’s licensed models are among the attractive options that companies can integrate into products. Model weights Home Hugging Face Sharing via the developers allows developers to create a working environment on their own servers. In addition, OpenAI compatible API servers are installed with infrastructures such as vLLM and SGLang. On the other hand, licensing conditions of the components that are based on the model should also be examined. It is necessary to look at the licensing obligations of the basic model rather than just looking at the license specified in the card. Ornith-1.5-122B aims to show that a powerful MoE model can be

    Etiketler:Ornith-1.5MoE mimarisi nedirOrnith-1.5 122B özellikleri nedirMoE mimarisi nedir
    Can Kepenekoğlu
    Can KepenekoğluWriter
    @CanKepenekoglu

    Comments (0)

    Log in to comment.

    Log in
    No comments yet. Be the first to comment!