Eclipse Theia 1.62 Release: News and Noteworthy

We are happy to announce the Eclipse Theia 1.62 release! The release contains in total 54 merged pull requests. In this article, we will highlight some selected improvements and provide an overview of the latest news around Theia.

For those new to Eclipse Theia, it is the next-generation platform for building IDEs and tools for the web or desktop, based on modern state-of-the-art web technologies. With Theia AI, part of the Theia platform, you can also build AI-powered tools and IDEs with ease. For more details about Eclipse Theia, please refer to this article and visit the Theia website.

The Theia project also releases a product, the Theia IDE. The Eclipse Theia IDE is a modern, AI-powered, and open IDE for cloud and desktop environments, aimed at end users. The Theia IDE is based on the Theia platform and also includes advanced AI powered features. For more details, see the Theia IDE website.

If you are looking for a simple way to check out the new release, please download and install the Theia IDE, which is based on Theia 1.62.

Eclipse Theia 1.62: Selected features and improvements

In the following, we will highlight some selected improvements in the new release. As usual, we cannot mention all 54 improvements, however we will focus on the most notable changes as well as changes visible to end users. The corresponding pull requests are linked under the respective heading when applicable.

Improved Support for Mistral Models

Theia 1.62 improves compatibility with Mistral models by addressing communication issues that previously affected usage through the OpenAI-compatible API. While Theia AI already allowed connecting to models via the OpenAI adapter, some users reported inconsistent behavior with Mistral-based services due to subtle differences in how certain API implementations interpret the protocol. Similar issues were seen earlier with other providers such as DeepSeek.

With these issues resolved in 1.62, Theia now supports Mistral models more reliably. The video below demonstrates using Theia Coder with a Mistral-hosted model to generate a basic calculator application from scratch.

This integration is particularly relevant given Mistral’s generous free-tier offering. While data submitted under the free plan may be used for training, this can still be an appealing option for open source contributors whose code is publicly available. Furthermore, Mistral’s latest model “Devstral” is available for self-hosting. As a European provider, Mistral may also be attractive for organizations that need to meet specific regional compliance requirements.

More details on Mistral’s API and offerings are available at: 👉 https://mistral.ai/products/la-plateforme.

Learn how to configure the Mistral models in the AI-powered Theia IDE:
👉 https://theia-ide.org/docs/user_ai/#mistral-models

Add Claude 4 Models to default list

Claude 4 Sonnet and Claude 4 Opus are supported in both Theia AI and the AI-powered Theia IDE. These recently released models from Anthropic are recognized for their advanced coding and reasoning capabilities. They can be used in Theia without requiring any updates—users can simply include them in their model configuration to get started.

Theia AI continues to support a wide range of LLMs from providers such as OpenAI, Anthropic, HuggingFace, Azure, or self-hosted deployments. The integration of Claude 4 Sonnet and Claude 4 Opus follows the same model-agnostic approach, enabling both tool builders and end users to adopt new models easily via configuration. In the AI-powered Theia IDE, end users can add these models themselves and use them with all available agents.

To streamline adoption, in the Theia 1.62 release, the Claude 4 models are being added to the default list of supported models in the Theia IDE. In the video below, we show how to manually configure the new models (for previous versions) and how it was used with Theia Coder to add it as a default.

Configuring Claude 4 models in Theia.

Please also see our first collected experiences with Claude 4.

Prompt Caching for Anthropic

Theia 1.62 introduces default token caching for Anthropic models, addressing a common issue encountered with large-context agent workflows. Particularly with Claude 4 and, to a lesser extent, Claude 3.7, prompt executions tend to consume a significant number of tokens—even for relatively conservative tasks—due to the models’ tendency to examine extensive context to produce high-quality responses. While this can be beneficial in terms of output accuracy, it increases costs and accelerates token limit exhaustion.

To mitigate this, token caching is now enabled by default for all Anthropic models. The caching mechanism marks system prompts, tool definitions, and previous messages for reuse in subsequent interactions. This is especially effective for typical agent workflows, such as those using @Coder, where repeated or static prompt elements are common. By avoiding redundant token computation, users benefit from significantly reduced token usage and lower cost per interaction.

Token tracking in Theia has also been updated to reflect cache usage, ensuring that developers have an accurate understanding of actual token costs. In the screenshot below, you can see that for typical file interactions, most tokens can actually be cached.

Cached tokens are priced differently: writing to the cache costs 1.25× the base input token price, but reading from cache costs only 0.1×. Additionally, tokens retrieved from the cache are not counted against rate limits, which helps avoid throttling in high-throughput workflows. In most scenarios, this will lead to significantly lower token usage and costs.

Introduce experimental generic LLM provider via Vercel AI

Theia 1.62 introduces an experimental integration with Vercel AI as a generic large language model (LLM) provider. Theia AI already maintains a broad compatibility with a wide range of LLM backends. From cloud-hosted services like OpenAI, Anthropic, and Azure to self-hosted options like Ollama and LM Studio, Theia AI is designed to remain vendor-agnostic and adaptable. This approach ensures that development workflows are not tightly coupled to any specific model or provider. For more details, also see the article Why Theia Supports Any LLM.

The Vercel AI integration supports this vision by offering a unified, generic interface for accessing multiple providers through the Vercel AI SDK. This allows Theia to explore the feasibility of minimizing the need for maintaining separate code paths for each provider—an increasingly relevant concern as the number of LLM options grows. The implementation already supports key features such as tool calling, streaming, structured output, and token usage tracking, and has been tested with a variety of OpenAI and Anthropic models.

Out of the box, users can try common OpenAI and Claude variants. Configuration of new models is straightforward via Theia preferences.

👉 See the Theia AI documentation for Vercel AI for more details

We encourage users and adopters to actively test this Vercel-based provider and provide feedback. Using it as an alternative to existing integrations not only helps reduce long-term complexity, but also contributes to stabilizing and shaping this generic interface for broader future use.

Prompt management refactoring and view

Theia 1.62 introduces a refactored system for managing AI prompt fragments and system prompts, along with a new configuration view. This updated view allows users to see all defined system prompts, including their associated agents, variants, and any applied customizations. Users can explore which variants are selected or set as default and can manage customizations directly from the interface, including reverting or deleting them when necessary (see video below).

New prompt management interface showing configuration options.

The refactoring also clarifies the distinction between prompt templates and prompt fragments. Prompt templates represent the raw text, while prompt fragments combine these templates with metadata. Each prompt fragment can now have a built-in version (at most one) and multiple custom versions. Prompt management is centered around PromptVariantSets, which group one or more fragments under a shared ID, allowing agents to use sets of prompt variants instead of individual templates.

The updated prompt service handles all prompt-related logic, including system prompts and user-defined customizations. This consolidation improves state handling and decouples prompt customization from other services. Unassociated prompt fragments are also shown in the configuration interface, making it easier to manage incomplete or experimental fragments. The refactoring is designed to maintain compatibility with existing workflows while supporting more flexible and scalable prompt configurations. We plan to extend on this initial improvement in future releases to ease the prompt management for users and adopters alike.

WSL remote support

Theia 1.62 adds support for remote workspaces via Windows Subsystem for Linux (WSL). This addition extends the existing remote functionality to include local WSL distributions, allowing users on Windows to interact with their Linux-based development setups more seamlessly.

When using Theia on Windows, users can now select WSL commands from the remote menu, accessible through the button in the lower-left corner. These commands allow users to connect to WSL distributions or directly reopen folders in these. Once a folder is reopened this way, it is also stored in the list of recent workspaces, making it easier to reconnect to the same environment in future sessions.
The Video below shows how to reopen a a local workspace in WSL:

Implement “code actions on save”

Theia 1.62 contains an important fix to enable code actions to be executed automatically upon saving a file. This functionality allows developers to configure specific actions—such as organizing imports—to trigger consistently when saving.

For instance, as shown in the screenshot below, when an import is added and deliberately placed incorrectly within the source code, the configured code action will reorder it appropriately the moment the file is saved.

Demonstration of automatic code actions executing on file save.

For Adopters: Refactor menu nodes

Theia 1.62 introduces active menu nodes through a new infrastructure that simplifies the handling of menu and toolbar items. Previously, menu nodes—such as actions, sub-menus, groups, and separators—were treated as static data, placing the burden on consumers to determine visibility, enablement, and parameter transformations based on the execution context.

With this update, menu nodes have been made polymorphic objects that can manage their own behavior. This includes evaluating whether they should be visible or enabled and adapting command parameters, such as translating widget instances to view identifiers. These capabilities are now encapsulated in the MenuCommandAdapterRegistry, which serves as a central mechanism to support advanced logic, particularly for VS Code plugin contributions.

By localizing complexity within the menu node implementation and plugin engine, this change reduces the need for special-case logic across the platform. It improves the maintainability of Theia’s command and menu system and paves the way for cleaner integration of external contributions.

As always, the 1.62 release contains much more than described in this article, e.g. the compatibility with VS Code Extensions has been upgraded to API version 1.99.3.
For a complete overview of all Theia AI updates in this release, please refer to the collecting epic for 1.62. All these features and improvements (in total 54) were the result of one month of intensive development. Eclipse Theia follows a monthly release schedule. We are looking forward to the next release due next month, stay tuned! To be notified about future releases, follow us on LinkedIn or follow Theia on Twitter and subscribe to our mailing list.

If you are interested in building custom tools or IDEs based on Eclipse Theia, EclipseSource provides consulting and implementation services for Eclipse Theia, for AI-powered tools, as well as for web-based tools in general.

Furthermore, if you want to extend Theia, Theia AI or the Theia IDE with new features or strategically invest into the project, EclipseSource provides sponsored development for Theia, too. Finally, we provide consulting and support for hosting web-based tools in the cloud.

👉 Get in contact with us, to discuss your use case!

👉 Follow us on LinkedIn!