# What changed in @ai-sdk/alibaba

> 14 changes in @ai-sdk/alibaba 0.0 to 2 from its official release notes, upgrade guides and advisories: 3 breaking, 2 new APIs and options. Each one cited. Sources checked 2026-10-18. HTML: https://unstale.sh/npm/@ai-sdk/alibaba/

To see which of these apply to a repository, run `npx @unstale/cli init` in it.

## @ai-sdk/alibaba 2

### Breaking changes

- Reasoning is preserved in multi-turn requests by default on supported models. (2.0.49) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.49)

### New APIs and options

- Adds max as a portable reasoning level, widening LanguageModelV4CallOptions['reasoning'] and the derived ReasoningLevel type. (2.0.62) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.62)
- Adds support for wan3 all-in-one video generation. (1.0.50, 2.0.35) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.35), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%401.0.50)

### Behavior and default changes

- Wan video requests use the Wan 2.7 wire format for later Wan 2 minors; future majors and unknown IDs use the current all-in-one media protocol instead of legacy size and reference fields. (1.0.63, 2.0.66) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.66), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%401.0.63)
- Top-level embedding dimensions are mapped to provider settings. (2.0.65) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.65)

### Fixes and other changes

- The User-Agent header is now standards-compliant. (2.0.58) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.58)
- Fixes streamed tool calls with missing, blank, or repeated IDs. (2.0.57) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.57)
- Uses model-specific structured output modes. (0.0.29, 1.0.57, 2.0.52) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.52), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%401.0.57), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%400.0.29)
- Prevents negative text output token counts when providers report reasoning tokens; Perplexity reasoning tokens are treated separately from completion tokens. (1.0.50, 2.0.35) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.35), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%401.0.50)
- Handle empty string tool call IDs. (2.0.33) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%402.0.33)

## @ai-sdk/alibaba 1

### Breaking changes

- Reasoning is preserved by default in multi-turn requests on supported models. (1.0.55) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%401.0.55)

### Fixes and other changes

- Reasoning streams are preserved across empty tool-call deltas for DeepSeek. (1.0.54) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%401.0.54)

## @ai-sdk/alibaba 0.0

### Breaking changes

- Reasoning is preserved by default in multi-turn requests on supported models. (0.0.28) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%400.0.28)

### Fixes and other changes

- Reasoning streams are preserved across empty tool-call deltas. (0.0.27) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/alibaba%400.0.27)