Optuna v5 Roadmap

More than five years have passed since the release v1.0 of Optuna, and its applications have expanded beyond machine learning hyperparameter optimization to encompass material discovery and diverse other fields. As a result, we continue to receive numerous requests for feature enhancements, and its use cases are continually growing.

We are planning to release Optuna v5.0 in the summer of 2026. This roadmap explains what the Optuna development team will prioritize in future development. Actual development will follow the same approach as previous Optuna development, with regular minor releases, and new features, services, and products will be released as they are completed.

This roadmap serves three purposes:

• To inform existing Optuna users about how the tool’s capabilities will be enhanced and expanded to support new use cases.
• To introduce those who have yet to use Optuna to the novel possibilities that black-box optimization offers.
• To provide guidelines for those interested in joining the Optuna development community on how to participate effectively.

Implemented features will be released incrementally, and we encourage you to try them out and provide feedback directly via GitHub issue. Future development plans are introduced on optuna.org. Those interested in contributing to the v5.0 roadmap are invited to fill out this form. Several development themes have been concretely defined, with well-structured tasks compiled in GitHub issues. Those wishing to participate should refer to these and submit your Pull request accordingly.

Core Concepts of Optuna v5.0

Optuna is being developed as a framework that provides efficient solutions for various black-box optimization problems and offers straightforward analysis tools. In Optuna v5.0, we aim to significantly expand the application domains of diverse black-box optimization techniques. We also seek to enhance usability for users who haven’t previously adopted Optuna or haven’t utilized its advanced features. This roadmap presents our development plan for v5.0, focusing on three key perspectives:

• Unleashing the potential of black-box optimization with generative AI: We will approach this in two ways: by constructing a black-box optimization toolchain for generative AI, and by implementing generative AI-based black-box optimization algorithms.
• Making Optuna available anytime, anywhere: For existing users, we will provide a streamlined setup with minimal configuration requirements, while simultaneously offering unified, cross-environment/language interfaces for new users.
• Continuing the sustainable growth of Optuna: We will stabilize and enhance existing features while introducing mechanisms to handle new problem domains that are not currently supported by Optuna.

Schedule

Below is the development schedule for Optuna v5.0. We will continue releasing minor updates every 2–3 months, gradually unveiling features as part of v5.0.

Unleashing the Potential of Black-box Optimization with Generative AI

Project Overview

In Optuna v5.0, we are planning to expand the possibilities of black-box optimization from both technical development for AI-generated content and technological development utilizing generative AI.

The development plan to achieve this goal includes, among others, the following two development themes:

• Development of a Toolchain for Prompt Optimization
• Development of a Toolchain for LLM Agent Systems

Development Themes

Development of a Toolchain for Prompt Optimization:

Prompt engineering is widely recognized as essential for maximizing LLM performance. Prompt optimization, which automates or semi-automates this process, has recently attracted significant attention, with numerous research and product developments underway. However, no de facto standard product currently exists in this field. We approach this problem by leveraging Optuna’s powerful algorithms, analytical capabilities, and experiment management features. Our solution aims to become broadly useful not only for black-box optimization but also for various LLM-based development and research applications

Development of a Toolchain for LLM Agent Systems:

Agent systems that independently select and utilize tools have recently garnered significant attention, potentially ushering in a new paradigm for product development. The Model Context Protocol (MCP), currently under development as open-source software, is emerging as a de facto standard. We plan to provide an MCP server for utilizing Optuna while also offering guidance for integrating with other MCP servers and typical MCP clients. This will enable users to perform valuable analyses and problem-specific improvements without needing to become proficient in individual tools.

Development Items

The development items associated with these themes, including those currently in progress or under formulation, are as follows:

Development of a Toolchain for Prompt Optimization:

• Development of a feature that suggests prompt candidates through Optuna’s interface
• Development of an intuitive user interface
• Development of functionality to support prompt evaluation

Development of a Toolchain for LLM Agent Systems:

• Implementation of an MCP server utilizing Optuna
• Development of an agent that automatically analyzes optimization history and provides improvement suggestions
• Establishment of guidelines for integration with other MCP servers and typical MCP clients

Making Optuna Available Anytime, Anywhere

Project Overview

Optuna v5.0 aims to provide users who have not used Optuna’s advanced features, as well as to enable users who have never used Optuna before, such as those who do not usually write programs or use programming languages other than Python, to use Optuna.

The development plan to achieve this goal includes the following themes:

• Optimized Performance for Resource-constrained Environments
• Enable Optuna Usage in Non-Python Environments

Development Themes

Optimized Performance for Resource-constrained Environments:

While Optuna itself is purely written in Python with carefully designed implementation and carefully selected dependency packages, it has been determined that significantly reducing the software’s memory footprint from its current level remains challenging due to the inherent limitations of Python’s language specifications. Furthermore, while continuous efforts have been made over the years to improve performance, Python inherently has limitations in achieving further speed optimizations. Therefore, we are currently considering redesigning Optuna to include only essential functionality and re-implementing it in Rust to address these issues. This high-performance, low-memory Optuna implementation using Rust is intended for use cases requiring embedded systems or handling large numbers of trials.

Enable Optuna Usage in Non-Python Environments:

While Optuna currently provides Python API and CLI interfaces, we recognize that there are significant barriers preventing non-programmers or users of non-Python programming languages from utilizing it. To make Optuna’s utility accessible to users who don’t write code at all, we are considering implementing mechanisms that enable black-box optimization using Optuna directly in Google Spreadsheets, as well as systems that automatically execute and analyze black-box optimization by processing problems described in natural language. Furthermore, we have a plan to provide multi-language support beyond Python by compiling a Rust reimplementation of Optuna into WASM.

Development Items

The development items associated with these themes, including those currently in progress or under development, are as follows:

Optimized Performance for Resource-constrained Environments:

• Prototyping redesign of Optuna and Rust reimplementation
• Evaluating performance benefits from reimplementing specific components in Rust
• Exploring deployment and operational maintenance models for partial Rust implementations of Optuna

Enable Optuna Usage in Non-Python Environments:

• Prototyping templates that enable black-box optimization from Google Spreadsheets
• Achieving multi-language compatibility through WASM conversion of Rust implementations

Continuing the Sustainable Growth of Optuna

Project Overview

Optuna v5.0 aims to stabilize and enhance the performance of many features currently widely used by existing users, while also introducing mechanisms to handle new problem domains that Optuna currently doesn’t cover.

This goal encompasses various types of development themes, which can be broadly categorized into three main frameworks:

• Implementing High-impact Improvements for the Majority of Users
• Delivering High-impact Improvements for Specific Use Cases
• Introducing New Approaches to Cover a Wider Range of Problem Settings

Development Themes

Implementing High-impact Improvements for the Majority of Users:

We aim significantly to enhance the Optuna experience for many users through improvements to the default sampler, enhanced AutoSampler capabilities for automatic algorithm switching, strengthened benchmarking guidance for sampler selection, and substantial enhancements to Optuna’s experiment management functionality.

Delivering High-impact Improvements for Specific Use Cases:

Optuna supports multi-objective optimization problems and constrained optimization problems, frequently employed in specific applications such as neural architecture search and material design. Implementing specialized improvements tailored to these particular use cases can generate profoundly significant impacts. We approach this challenge by developing: a Gaussian process-based multi-objective optimization algorithm that excels in scenarios with limited evaluation iterations, and constrained optimization algorithms for cases where specific parameter evaluations are unfeasible or constraints are specified through explicit linear inequalities.

Introducing New Approaches to Cover a Wider Range of Problem Settings:

Some problem domains that Optuna currently cannot handle include cutting-edge research areas that have recently gained significant attention in the field of black-box optimization. Developing frameworks to address such problems could not only help disseminate the latest research findings to the community but also attract talented new contributors. We are particularly considering implementing support for new problem domains such as Multi-fidelity optimization and Exploratory Landscape Analysis (ELA) within Optuna.

Development Items

The development items associated with these themes, including those currently in progress or under formulation, are as follows:

Implementing High-impact Improvements for the Majority of Users:

• Enhancements to the default sampler across various configuration settings
• Feature upgrades for AutoSampler to enable multi-objective optimization and constrained optimization with automatic algorithm selection

Delivering High-impact Improvements for Specific Use Cases:

• Introduction of a Gaussian process-based multi-objective optimization algorithm
• Support for constrained optimization accounting for cases where evaluation of certain parameters is infeasible

Introducing New Approaches to Cover a Wider Range of Problem Settings:

• Introduction of a framework for handling multi-fidelity optimization
• Introduction of integration with third-party products supporting ELA (Expected Learning Algorithm)

Conclusion

The Optuna v5.0 roadmap outlines our development strategy moving forward. We will focus on three core initiatives: (1) expanding the possibilities of black-box optimization through generative AI, (2) making Optuna always accessible and convenient to use, and (3) continuously advancing Optuna’s capabilities.

Implemented features will be released incrementally, and we encourage you to try them out and provide feedback directly via GitHub issue. Future development plans are introduced on optuna.org. Those interested in contributing to the v5.0 roadmap are invited to fill out this form. Several development themes have been concretely defined, with well-structured tasks compiled in GitHub issues. Those wishing to participate should refer to these and submit your Pull request accordingly.