Skip to content
  • There are no suggestions because the search field is empty.

2025 Changelog

Oct 10th 2025

Version 4.4

Compass API

Grok 2.5 (Preview)
A large-scale, multimodal model developed by xAI that uses a Mixture-of-Experts (MoE) architecture.

Web Search
Web Search enables users to retrieve search engine results based on user queries in a single API call. The search results include the frequently clicked links on the destination website. Supported with GPT-4.1, GPT-4o, and GPT-4o mini models.

Compass Portal

Request New Models
Users can now easily request new models not listed in the Model Catalog page by clicking Request Model button available on the page and submitting the request. This enhancement ensures users can quickly notify Compass Support when they are unable to find a specific model.

Sept 30th 2025
Version 4.3

Compass API

gpt-realtime
gpt-realtime is a realtime model capable of responding to audio and text inputs in realtime.

Llama 3.2 90B Vision
The Llama 3.2-Vision is a collection of pretrained and instruction-tuned image reasoning generative models.

Cohere Command R+
Command R+ is a large language model optimized for conversational interaction and long-context tasks.

o3 Region Update
o3 model is now available in the UAE region and sovereign.

gpt-oss-120b Azure Model Retirement
The gpt-oss-120b-azure  model ID has been officially retired and is no longer available on the Compass Platform. Users are encouraged to transition to alternative model versions available on the Compass Platform, which include gpt-oss-120b-cerebras, gpt-oss-120b -core42, gpt-oss-120b-core42-amd, and gpt-oss-120b-qualcomm.

Response API Reference Update
Response API Reference now supports the o3 model and Function Calling functionality.

Compass Portal

Token & Latency Metrics in Chat Playground
Users can now view detailed token usage and response latency after every completion in the Chat Playground on the Compass portal. After each message is completed, information related to the total tokens consumed and response latency (completion time) is displayed, providing users with transparency into model usage. This enables them to monitor performance, manage token consumption, and improve operational efficiency with key performance data.

Activity Log Dashboard
Users can now access their activity directly via a new navigation entry in the admin panel. This new dashboard offers an intuitive interface for browsing all activity logs, providing clear visibility into platform actions. 

Real-Time Cost Visibility for Pay-As-You-Go Models
View real-time cost consumption for Pay-As-You-Go models, empowering better cost governance and operational efficiency. This enhancement provides immediate visibility into model usage costs across deployment types, with clarity and precision.


 
Sept 17th, 2025

Version 4.2.1

Compass API

K2 Think
K2 Think is an open-source reasoning model that achieves state-of-the-art performance with 32B parameters. It was developed in the UAE by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI).
- k2-think-core42: The model is deployed and running on the Core42 cloud located in the UAE region.
- k2-think-cerebras: The model is deployed and running on the Cerebras in the US region.

Aug 29th, 2025
Version 4.2

Compass API

GPT-5 (Preview)
An advanced model with enhanced understanding, reasoning ability, and designed to be more accurate, flexible, and efficient.

GPT-5 mini (Preview)
GPT-5 mini offers faster performance, making it ideal for well-defined tasks and clear, focused prompts.

GPT-4o Transcribe (Preview)
A speech-to-text model that utilizes GPT-4o to transcribe audio.

GPT-4o mini Transcribe (Preview)
A speech-to-text model that uses GPT-4o mini to transcribe audio. 

GPT-4o mini TTS (Preview)
A text-to-speech model built on GPT-4o mini is used to convert text to natural-sounding spoken text. 

Cohere Command A (Preview)
An advanced model offering enhanced performance across a variety of tasks.

Web Search API Retirement
The Web Search API has been officially retired and is no longer available on the Compass Platform.

Responses API – Expanded Model Support
Now supports GPT-4o mini, o4-mini, and o3-mini models.

File Search
File Search is now available via Responses, enabling grounded retrieval across uploaded documents. 

Fine-Tuning
Fine-tuning is now available for GPT-4.1. This capability is sovereign within the UAE. 


14th Aug 2025
Version 4.1

Compass API

gpt-oss-120b (Preview)
A highly flexible, open-source model by OpenAI licensed under Apache 2.0. The model is designed for reasoning and efficient performance. Compass offers the following model versions on AMD and Qualcomm, located in the UAE region:
- gpt-oss-120b-core42-amd: The model is deployed and running on AMD.
-gpt-oss-120b-qualcomm: The model is deployed and running on Qualcomm. 

    gpt-oss-20b (Preview)
    A highly flexible, open-source model by OpenAI licensed under Apache 2.0. This model is ideal for experimentation, customisation, and commercial deployment. The model is deployed and running on the AMD and Qualcomm located in the UAE region.
    - gpt-oss-20b-core42-amd: The model is deployed and running on AMD.
    - gpt-oss-20b-qualcomm: The model is deployed and running on Qualcomm.

    Cohere Embed 4 (Preview)
    Enables organizations to search their unstructured documents.

    DeepSeek R1 0528 (Preview)
    Offers enhanced reasoning capabilities.

    OpenAI Agent SDK: The Compass API now supports all Chat Completion models in the Agent SDK, except Mistral 7B and Mixtral 8x7B models.

    Response Endpoint: The Response Endpoint is now enabled on Compass API and supported with GPT-4o and GPT-4.1 models.

    Fine-Tuning (GPT-4.1)
    Fine-tuning for the GPT-4.1 model is now available on the Compass platform.

    08th Aug 2025
    Version 4.0.1

    Compass API

    gpt-oss-120b (Preview)
    A highly flexible, open-source model by OpenAI licensed under Apache 2.0. This model is optimised for high-performance tasks. The model is designed for reasoning, tool use, and efficient performance. Compass offers the following three model versions:
    – gpt-oss-120b Cerebras: The model is deployed and running on high-performance Cerebras clusters.
    – gpt-oss-120b Core42: The model is deployed and running on the Core42 cloud, located in the UAE region.
    – gpt-oss-120b Azure: The model is deployed and running on the Azure cloud located in the UAE region.

    gpt-oss-20b Core42 (Preview)
    A highly flexible, open-source model by OpenAI licensed under Apache 2.0. This model is ideal for experimentation, customisation, and commercial deployment. The model is deployed and running on the Core42 cloud located in the UAE region.

    o4-mini
    Engineered for compact efficiency, it delivers fast, cost-efficient reasoning with exceptional performance in maths, coding, and visual tasks. Ideal for developers and creators seeking speed without compromise.

    Qwen3-14B
    A 14.8B parameter model, optimised for multilingual reasoning, coding, and agentic tasks. Supports 119 languages, long context (up to 131k tokens), and hybrid reasoning modes for flexible performance.

    Qwen3-Embedding-8B
    A new series of multilingual embedding models optimised for text ranking, retrieval, classification, and code search. Supports over 100 languages, flexible embedding dimensions (up to 4096), and instruction-aware tuning.

    Qwen3-Reranker-8B
    An 8B parameter model built to enhance search quality by reranking documents based on query relevance. Supports over 100 languages, understands long texts with a 32k context length, and delivers state-of-the-art performance in text and code retrieval tasks.

    Fine-Tuning (GPT-4o)
    Fine-tuning for the GPT-4o model is available on the Compass platform. You can now fine-tune GPT-4o to customise model behaviour, tone, and domain-specific performance. This supports multilingual use cases with enhanced accuracy and efficiency.

    31st July 2025
    Version 4.0

    Compass API

    Claude Sonnet 4 (Preview)
    A high-performance reasoning model, particularly strong in coding tasks. Ideal for applications ranging from virtual assistants to enterprise-scale operations. Currently available to select customers.

    Cohere Command R (Preview)
    An advanced conversational AI model engineered for handling long-form, natural language interactions. Currently available to select customers.

    Compass Portal

    Department Management
    Users can now create and manage departments directly from their profiles. Each department can contain multiple projects, with assigned resources and API keys visible on the project details page.

    API Key Management Enhancements
    A new filtering capability has been added to the API Keys section. Users can now quickly sort keys by department, project, and environment—simplifying navigation in large environments.

    Portal URL Change
    The Compass Portal is now accessible at: https://compass.core42.ai
    The legacy interface has been fully retired.

    19th Jun
    Version 3.7

    Compass Portal
    Compass is excited to announce that a new version of Compass platform is now live! Users can explore and preview new features. Dive in to experience the refreshed user interface and discover upcoming functionalities.

    31st May
    Version 3.5

    Compass API
    GPT Image 1:
    Compass introduces the GPT Image 1 model, an image generation and natively multimodal model that accepts text and image inputs and produces image outputs.

    o3:
    Compass introduces the o3, a versatile and robust model that sets a new benchmark for math, science, coding, and visual reasoning. This model excels in technical writing and instruction-following and is capable of solving complex multi-step problems that involve text, code, and image analysis.

    11th May
    Version 3.4

    Compass API
    o3-mini:
    Compass introduces o3-mini, the latest compact reasoning model, offering advanced intelligence including essential developer capabilities such as structured outputs and function calling.

    o1:
    Compass introduces the o1 model, designed for complex reasoning tasks, and processing information in a multi-step manner to handle intricate prompts. Its architecture allows for multimodal inputs, enhancing its ability to understand text and images, resulting in coherent text output.

    Model Deprecation:
    The gpt-4o-realtime-preview-2024-10-01 model has been deprecated and is no longer available. Suggested replacement: gpt-4o-realtime-preview (version gpt-4o-realtime-preview-2024-12-17).

    29th Apr
    Version 3.3

    Compass API
    GPT-4.1 (Preview):
    Compass introduces GPT-4.1, a multimodal model for complex tasks and well-suited for problem-solving across domains. It offers increased efficiency for faster processing and a streamlined user experience.

    09th Apr
    Version 3.2.4

    Compass API
    GPT-4o Audio (Preview):
    Compass introduces GPT-4o Audio model that provides real-time, natural voice interactions with improved speed, emotion, and responsiveness for a seamless conversational AI experience.

    02nd Feb
    Version 3.1

    Compass Portal
    Batch Jobs Dashboard:
    We have released a new dashboard that offers an intuitive interface, allowing users to create and manage batch jobs with just a few clicks.

    Set Spend Limit for Batch Jobs:
    Users can set separate daily and monthly spend limits for online and batch processing on supported Pay-as-You-Go models.

    Compass API
    Llama 3.3 70B (Preview):
    We have featured the latest version of the Llama model, Llama 3.3 70B. This enhanced, high-performance model is both cost-effective and versatile, supporting a wide range of use cases. Now available via the Compass API.

    Batch Processing:
    Now supports GPT-4o mini.

    Web Search:
    Now supports Mistral 7B and Mixtral 8x7B.