Skip to main content

AI-Starter 2.4.15

· 6 min read
Takaaki Kakei
Takaaki Kakei
Owner

We are pleased to announce the release of the latest version of AI-Starter, version 2.4.15. This update includes more changes than usual.

Main Updates​

  1. Models
  2. Image Generation Models
  3. UI/UX Improvements
  4. Feature Additions and Enhancements
  5. Other Minor Bug Fixes and Improvements

1. Models​

We have added and updated the following models:

  • AWS
    • Removed conversation limits for Claude 3.7 Sonnet in Extended Thinking Mode
    • Added support for Nova Premier
  • Azure
    • Added support for o4-mini
    • Added support for GPT-4.1
  • Google
    • Added support for Gemini 2.5 Flash Preview
    • Added support for Gemini 2.5 Pro Preview
    • Updated model ID for Gemini 2.0 Flash to an alias
    • Updated model ID for Gemini 2.0 Flash-Lite to an alias
  • OpenAI
    • Added support for o4-mini
    • Added support for o3
    • Added support for GPT-4.1
    • Added support for GPT-4.1 Mini
    • Added support for GPT-4.1 Nano
    • Removed "Confidential information cannot be entered" restriction for GPT-4o Search Preview
    • Removed "Confidential information cannot be entered" restriction for GPT-4o mini Search Preview

If you wish to add or update models, configuration changes may be required, so please contact our support desk. For information handling in OpenAI's "GPT-4o Search Preview," please also contact our support desk. Please note that for the Nova series, we have confirmed an issue on AWS where hallucinations may occur when inputting PDFs.

2. Image Generation Models​

The image generation infrastructure of AI-Starter has been significantly improved. In this update, the following image generation models have been added or updated:

  • AWS
    • Added support for Nova Canvas
  • Azure
    • Added support for DALL-E 3
  • Google
    • Added support for Imagen 3
  • OpenAI
    • Added support for GPT Image 1

Nova Canvas is an image generation model in the Nova series. In this update, only text-to-image generation is supported. Please enable and use the model in the Tokyo region.

https://docs.aws.amazon.com/ja_jp/nova/latest/userguide/prompting-image-generation.html

Previously, DALL-E 3 was only supported via OpenAI in AI-Starter, but now it is also supported via Azure OpenAI.

https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/dall-e?tabs=gpt-image-1

Google's Imagen 3 is an image generation model from Google. It can generate high-quality images at relatively high speed. Due to language restrictions, instructions must be given in English.

https://cloud.google.com/vertex-ai/generative-ai/docs/image/overview

OpenAI's GPT Image 1 is a recently released image generation model. Unlike the models above, it allows you to edit images generated in the recent past. Support for GPT Image 1 in Azure OpenAI is planned once operation is confirmed.

https://openai.com/index/image-generation-api/

The AI-Starter development team's overall assessment is as follows:

Comparison of Image Generation ModelsComparison of Image Generation Models
  • Imagen 3 offers the best balance
  • GPT Image 1 is expensive but allows high-quality, interactive image generation
  • Nova Canvas is low-cost and fast

3. UI/UX Improvements​

As announced in Upcoming Features in AI-Starter 2.4.15, we have made improvements to the UI/UX. The sidebar structure has been redesigned for more intuitive operation. The new structure is designed to flexibly accommodate future feature additions and expansions. A new sidebar toggle feature has also been added, allowing users to adjust the UI according to their usage scenarios and preferences. Additionally, minor UI improvements such as changes to prompt template icons have been made.

New Sidebar ScreenNew Sidebar Screen New Sidebar Screen (Collapsed)New Sidebar Screen (Collapsed)

4. Feature Additions and Enhancements​

  • Support for Bedrock prompt caching
  • Addition of BedrockKnowledgeBasesRetrieval option
  • Addition of GoogleDriveFiles option (Beta)
  • Addition of OpenAIResponsesAPI option

Bedrock prompt caching is now supported. This feature enables reduced response times and cost savings for some Bedrock models. It is automatically applied within AI-Starter, so administrators do not need to change any settings.

https://docs.aws.amazon.com/ja_jp/bedrock/latest/userguide/prompt-caching.html

The BedrockKnowledgeBasesRetrieval option has been added. This allows you to perform searches in Bedrock Knowledge Bases and implement RAG (Retrieval-Augmented Generation) with any model. Features such as reranking and filtering are also supported to improve the accuracy and flexibility of RAG.

A Google Drive Retrieval option (Beta) has been added. This allows you to incorporate the contents of specified Google Drive files into conversations. Currently, the following file types can be imported:

  • Google Docs
  • The first sheet of Google Sheets

The OpenAIResponsesAPI option has been added. The Responses API is a new API for building agents using OpenAI's built-in tools. Currently, two tools are supported: web search and file search. Additional tools such as code interpreter and MCP are planned for the future. Please use this if you want to easily build agents.

https://openai.com/index/new-tools-for-building-agents/

5. Other Minor Bug Fixes and Improvements​

Other minor bug fixes and improvements have been made. At the same time, some models and options are no longer supported.

The following models are no longer supported. Assistants using these models can still be used, but issues may arise with future updates. We recommend migrating to newer models with better performance and cost efficiency.

  • Azure
    • o1-mini (replaced by the superior o4-mini)
    • o3-mini (replaced by the superior o4-mini)
    • GPT-4.5 Preview (scheduled for deprecation; replaced by GPT-4.1)
  • Google
    • Gemini 2.0 Flash-Lite Preview (replaced by GA version of Gemini 2.0 Flash-Lite)
    • Gemini 2.0 Pro Experimental (discontinued)
    • Gemini 1.5 Pro (replaced by the superior Gemini 2.0 series)
  • OpenAI
    • GPT-4.5 Preview (scheduled for deprecation; replaced by GPT-4.1)
    • o1-mini (replaced by the superior o4-mini)
    • o3-mini (replaced by the superior o4-mini)
    • o1 (replaced by the superior o3)

The following options are no longer supported:

  • BingSearchRetrieval (replaced by OpenAI models with web search functionality)

Conclusion​

Thank you for reading to the end. AI-Starter is committed to continuous improvement, driven by feedback from our users. Please review the contents of this update, and if you have any comments or requests, we would love to hear from you. Your feedback will help AI-Starter evolve even further. Please look forward to future updates.