Kevin Jain

This commit performs a **maintenance update** to the **AI model type definitions**, specifically within the `types/generated-snapshot/experimental/index.ts` and `types/generated-snapshot/latest/index.ts` files. It **adds support for several new AI models**, including `@cf/moonshotai/kimi-k2.5`, `@cf/zai-org/glm-4.7-flash`, `@cf/nvidia/nemotron-3-120b-a12b`, and multiple `@cf/black-forest-labs/flux-2` variants. This ensures that developers can now interact with these newly integrated models with **type safety**, improving code reliability and development experience across the platform by providing accurate and up-to-date interfaces.

Refactors the internal Cloudflare AI API module (`ai-api.ts`) by introducing `instanceof` checks and making the `tags` field optional, which may impact how consumers process AI API responses. Includes minor type fixes for consistency.

This change introduces a new capability for Cloudflare's internal AI bindings, allowing them to accept `FormData` as input. This enables integration with AI models that require multi-part data, such as those handling multiple image references.

This commit enhances Workers AI by adding type support for OpenAI's Responses API format and integrating several new AI models. This expands the capabilities of Workers AI, allowing it to correctly interact with and utilize these new models and response structures.

This update to Workers AI types introduces support for several new AI models and a `websocket` option in `AiOptions`, significantly expanding the range of AI capabilities available to developers using the platform.

This commit introduces a crucial **bug fix** to the **Workers AI UI components** (`ModelBadges.tsx`, `ModelFeatures.tsx`) by correcting the timestamp calculation for planned model deprecation dates from seconds to milliseconds, ensuring accurate display to users. Concurrently, it performs a substantial **maintenance update** by adding planned deprecation dates to over twenty **Workers AI models** through updates to their respective JSON configuration files. Furthermore, the commit includes **documentation updates** to the changelog and release notes, formally announcing the introduction of the `EmbeddingGemma` model. This work enhances the accuracy of model lifecycle information and improves user awareness across the **Workers AI platform**.

This commit **enhances the Workers AI platform** by **adding support for a new model** and updating existing model schemas. It introduces the `embeddinggemma-300m` model with its full configuration, including description, task, and input/output schema, making it available for use. Additionally, this **feature enhancement** updates the schemas for `lucid-origin`, `nova-3`, and `smart-turn-v2` models, refining input parameters, renaming fields, and adding new properties like `realtime` and `steps`. This ensures the **Workers AI model catalog** documentation is accurate and reflects expanded capabilities for various AI tasks.

This is an isolated change primarily affecting internal AI API tests and a related internal API file. It has no production impact.

Isolated refactoring within the `cloudflare/internal/ai-api` module; no functional changes or external impact.

This change introduces a significant new capability to the Workers AI binding, allowing users to establish real-time, bi-directional WebSocket connections with supported AI models. This enables new interactive inference use cases for all Workers AI users.

This commit introduces a **new capability** to the **Workers AI binding**, enabling it to accept `ReadableStream` objects as inputs for specific models. This **feature enhancement**, primarily implemented in `src/cloudflare/internal/ai-api.ts`, allows users to send larger media files, which is expected to result in faster inference and unlock new possibilities for future AI models. The change is designed to be **backwards compatible** and is thoroughly validated with new test cases in `src/cloudflare/internal/test/ai/ai-api-test.js` and an updated mock in `src/cloudflare/internal/test/ai/ai-mock.js`.

This commit **enhances the AI integration** by **adding new AI model type definitions** to the `types/generated-snapshot` directories. It updates the `AiModels` interface and modifies the `Ai` class methods `autorag` and `run` within the **AI module**. These changes introduce support for critical new features, specifically enabling **asynchronous batch requests** and **streamable responses** for AI model interactions. This is a significant **enhancement** to the **AI capabilities**, allowing for more efficient and flexible interactions with various AI models.

Update Workers AI types

This commit **corrects the pricing information** for the **`@cf/myshell-ai/melotts` model** within the **Workers AI platform documentation**. The pricing unit, previously listed incorrectly as per input tokens, has been updated to accurately reflect **per audio minute**. This **documentation fix** ensures users consulting the `src/content/docs/workers-ai/platform/pricing.mdx` page have the correct cost structure for this specific AI model. Additionally, a new entry has been added to the **Workers AI release notes** (`src/content/release-notes/workers-ai.yaml`) to document this important correction.

This commit **introduces support for three new Workers AI models** by updating their type definitions across various API snapshots. It specifically adds type definitions for `@cf/myshell-ai/melotts`, `@cf/baai/bge-m3`, and `@cf/baai/bge-reranker-base`. This **new capability** enhances the **Workers AI** platform by making these models discoverable and type-safe for developers. The changes primarily affect the **generated type definitions** within the `types/generated-snapshot/` directory, providing improved developer experience through better type checking and auto-completion when interacting with these new models.

This commit provides a **bug fix** by correcting the **type definition** for the `segments` property within the **`whisper-turbo` model's output**. Previously, the type incorrectly represented `segments`; it is now accurately defined as an array across all relevant **generated type snapshots**. This ensures **type safety** and improves the **developer experience** for consumers of the `whisper-turbo` model, preventing potential type mismatches or runtime errors when accessing the `segments` data.

This commit **introduces new type definitions** across various **API snapshot versions** within the AI platform. It adds comprehensive type support for the **`@cf/meta/llama-guard-3-8b` model**, enabling developers to interact with this new AI capability in a type-safe manner. Furthermore, it **enhances existing text generation types** to include a `response_format` option, providing more control over the output structure. This **feature addition** ensures that client applications can leverage the latest AI models and features with improved code reliability and discoverability.

This commit **refactors** the **AI API client** by **removing a redundant return statement** within `src/cloudflare/internal/ai-api.ts`. It specifically addresses an unnecessary explicit return of the response body when handling `input.stream`, as this was already covered by a subsequent, unconditional return. This **code cleanup** improves the **readability and maintainability** of the **streaming response logic** without altering its functional behavior or introducing any breaking changes.

This commit performs **maintenance** by **updating the generated type definitions** for the **Workers AI** platform. It specifically adds type support for two new AI models: `@cf/deepseek-ai/deepseek-r1-distill-qwen-32b` and `@cf/unum/uform-gen2-qwen-500m`. This **chore** ensures that developers leveraging Workers AI can now use these models with proper type checking and autocompletion, improving code reliability and developer experience. The changes are reflected across various snapshot versions of the type definitions within the `types/generated-snapshot/` directory.

This commit **updates and refactors the type definitions** for the **Workers AI module**, primarily focusing on its supported AI models. It **adds new model specifications** to the generated type snapshots, including `whisper`, `llama-3.2-11b-vision-instruct`, and `flux-1-schnell`, while also converting the `AI.run()` function to be generic for improved flexibility. This **maintenance and feature enhancement** work ensures developers have **up-to-date and comprehensive type support** for the expanding Workers AI model catalog. The changes also prepare for the future migration of model-specific types to the new `@cloudflare/ai-types` NPM package.