feat: Add AI SDK v6 support by pablof7z · Pull Request #307 · OpenRouterTeam/ai-sdk-provider

pablof7z · 2025-12-17T23:46:16Z

Summary

This PR adds support for AI SDK v6 (currently in beta).

Changes

Migrate from ProviderV2 to ProviderV3
Migrate from LanguageModelV2 to LanguageModelV3
Migrate from EmbeddingModelV2 to EmbeddingModelV3
Update specificationVersion to 'v3'

Update usage structure to nested v6 format:

{
  inputTokens: { total, noCache, cacheRead, cacheWrite },
  outputTokens: { total, text, reasoning }
}

Update all type imports (SharedV2* -> SharedV3*, etc.)
Update dependencies:
- @ai-sdk/provider: 2.0.0 -> 3.0.0-beta.27
- @ai-sdk/provider-utils: 3.0.1 -> 4.0.0-beta.53
- ai peer dep: ^6.0.0

Testing

Tested with free models on OpenRouter:

✅ Basic text generation
✅ Streaming
✅ Tool calling
✅ Reasoning tokens (with DeepSeek R1 chimera)
✅ Multi-turn conversations

Notes

AI SDK v6 is currently in beta. This PR tracks the beta versions. Once v6 is stable, dependency versions can be updated to stable releases.

- Update package dependencies to AI SDK v6 beta versions - Migrate ProviderV2 → ProviderV3 - Migrate LanguageModelV2 → LanguageModelV3 - Migrate EmbeddingModelV2 → EmbeddingModelV3 - Update all type imports to V3 equivalents - Add comprehensive implementation documentation Known issues: - Usage structures need migration to V3 nested format - Test files need V3 type updates - Build currently fails due to usage type mismatches See AI-SDK-V6-IMPLEMENTATION.md for full details

- Migrate from ProviderV2 to ProviderV3 - Migrate from LanguageModelV2 to LanguageModelV3 - Migrate from EmbeddingModelV2 to EmbeddingModelV3 - Update specificationVersion to 'v3' - Update usage structure to nested v6 format: - inputTokens: { total, noCache, cacheRead, cacheWrite } - outputTokens: { total, text, reasoning } - Update all type imports (SharedV2* -> SharedV3*, etc.) - Update dependencies: - @ai-sdk/provider: 2.0.0 -> 3.0.0-beta.27 - @ai-sdk/provider-utils: 3.0.1 -> 4.0.0-beta.53 - ai peer dep: ^6.0.0 Tested with free models on OpenRouter: - Basic text generation ✅ - Streaming ✅ - Tool calling ✅ - Reasoning tokens ✅ - Multi-turn conversations ✅

socket-security · 2025-12-17T23:56:12Z

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Package	Supply Chain Security	Quality	Maintenance
@ai-sdk/provider@2.0.0 ⏵ 3.0.0-beta.27		⁺¹	^-1
@ai-sdk/provider-utils@3.0.18 ⏵ 4.0.0-beta.53	⁺¹		⁺¹
msw@2.12.4
ai@5.0.104 ⏵ 6.0.0-beta.159	⁺¹		⁺¹

View full report

- Remove documentation files from PR - Fix all V2→V3 type migrations in source files - Add MSW-based test server utility to replace removed createTestServer - Migrate all test files to V3 types - Fix JSONSchema7 type issues in tests - Fix requestBody unknown type issues - Typecheck now passes with 0 errors - Build succeeds - Live tests pass (basic, streaming, tool calling, reasoning, multi-turn)

src/tests/usage-accounting.test.ts

AviVahl · 2025-12-23T09:56:49Z

final v6 is released. would be great if this got updated, reviewed, merged, and released.

seannetlife · 2025-12-23T12:55:26Z

would this constitute a breaking change, or are the types backwards compatible?

elijah629 · 2025-12-24T17:16:18Z

https://ai-sdk.dev/docs/migration-guides/migration-guide-6-0 check here, I think they are compatible.

src/test-utils/test-server.ts

saeta-eth · 2025-12-26T21:10:49Z

would be great to see this PR move forward so we can start using AI SDK v6

ravisojitra · 2025-12-28T09:30:58Z

Any update on this? Relying totally on open router right now. and using v6 at the same time.

ZenAlexa · 2025-12-28T11:05:46Z

Just migrated our app to AI SDK v6 using this PR👍

- Update @ai-sdk/provider from 3.0.0-beta.27 to 3.0.0 - Update @ai-sdk/provider-utils from 4.0.0-beta.53 to 4.0.1 - Update ai from 6.0.0-beta.159 to 6.0.3 - Update tests for v6 usage format (nested inputTokens/outputTokens) - Update tests for v3 embedding specification version - Fix header tests for extended user-agent format

pablof7z · 2025-12-28T21:40:17Z

Updated to stable AI SDK v6 releases

I've updated the dependencies to use stable versions since v6 is now released:

@ai-sdk/provider: 3.0.0-beta.27 → 3.0.0
@ai-sdk/provider-utils: 4.0.0-beta.53 → 4.0.1
ai (dev): 6.0.0-beta.159 → 6.0.3

All tests pass with the stable versions.

src/tests/provider-options.test.ts

The-Best-Codes

Really hoping this can get finished up & merged soon! Would love to use OpenRouter with AI SDK in my applications! 🙌

wong2 · 2025-12-29T03:07:48Z

Yeah this is what's blocking our AI SDK v6 migration.

ldriss · 2025-12-29T22:41:02Z

+1

fwang2002 · 2025-12-30T03:13:27Z

There is asProviderV3 and asLanguageModelV3 in ai-sdk v6, but they are not exported. Copy these two files may help before ai-sdk-provider is ready.

vzt7 · 2025-12-31T03:07:18Z

Any update? 🙏

ldriss · 2025-12-31T11:29:49Z

No updates as user with enterprise plan i really need this v6 for agents and here we are i know the team is doing a lot of efforts, but 2 weeks and no single commit or update here is crazy

mattapperson · 2026-01-01T00:10:04Z

Tested locally with AI SDK v6. Found a critical issue:

LanguageModelV3FinishReason type mismatch

In AI SDK v6, finishReason is now an object { unified, raw } instead of a string. The PR is still returning strings like 'stop' and 'tool-calls'.

once this is fixed we can merge and work on a release. Thanks for this!

ldriss · 2026-01-01T00:28:50Z

Thank you @mattapperson

In AI SDK v6, the LanguageModelV3FinishReason type changed from a string to an object with `unified` and `raw` properties. This change ensures the provider returns the correct format: - `unified`: The normalized finish reason (stop, length, tool-calls, etc.) - `raw`: The original finish reason string from the API Changes: - Update mapOpenRouterFinishReason to return { unified, raw } object - Add createFinishReason helper for creating finish reasons - Update chat and completion streaming to use new format - Update all tests to expect object format

pablof7z · 2026-01-01T12:47:10Z

Fixed: LanguageModelV3FinishReason type

Thanks @mattapperson for catching this issue!

The finishReason is now correctly returned as an object with { unified, raw } format as required by AI SDK v6:

{
  unified: 'stop' | 'length' | 'content-filter' | 'tool-calls' | 'error' | 'other',
  raw: string | undefined  // Original finish reason from OpenRouter
}

All tests pass with the stable AI SDK v6 releases.

ldriss · 2026-01-01T12:54:41Z

Thanks @pablof7z

nvti · 2026-01-04T15:58:16Z

Any update? 🙏

wlib · 2026-01-04T16:42:21Z

@nvti Maintainers likely are on vacation mode until tomorrow, so I wouldn't expect to see anything until at least then.

ldriss · 2026-01-05T15:33:00Z

@subtleGradient Thank you so much

The-Best-Codes · 2026-01-05T15:40:39Z

Thank youuuu!!

haowang1013 · 2026-01-07T00:45:47Z

Any update?

ldriss · 2026-01-07T11:23:29Z

subtleGradient · 2026-01-07T17:20:26Z

great minds think alike
sorry I've been incommunicado about the progress on v6
will be working on the whole communication skills thing

but I have a gift for you:

https://github.com/OpenRouterTeam/ai-sdk-provider/releases/tag/v6.0.0-alpha.1

ldriss · 2026-01-07T17:22:06Z

@subtleGradient LOVE YOUU LOL

ZenAlexa · 2026-01-07T17:23:46Z

@subtleGradient Thank you so much! Happy 2026❤️

cpakken · 2026-01-07T17:59:01Z

great minds think alike sorry I've been incommunicado about the progress on v6 will be working on the whole communication skills thing

but I have a gift for you:

https://github.com/OpenRouterTeam/ai-sdk-provider/releases/tag/v6.0.0-alpha.1

Thanks, but having trouble enabling reasoning options. google/gemini-3-flash-preview no longer has reasoning even when I include

model: openrouter('google/gemini-3-flash-preview', {
      usage: {
        include: true,
      },
      reasoning: {
        enabled: true,
        effort: 'medium',
      },
    })
    ```

The-Best-Codes · 2026-01-07T19:16:54Z

@subtleGradient Thank you so much!!!

package.json

- Bump version to 2.0.0 - Fix zod peer dependency (^3.25.0 || ^4.0.0) - Make file content part fields optional and add file_id - Handle LanguageModelV3ToolApprovalResponsePart in tool responses - Apply biome formatting fixes

robert-j-y · 2026-01-23T22:34:32Z

Thank you @pablof7z and everyone else for your help on this!

* feat: Initial AI SDK v6 support implementation - Update package dependencies to AI SDK v6 beta versions - Migrate ProviderV2 → ProviderV3 - Migrate LanguageModelV2 → LanguageModelV3 - Migrate EmbeddingModelV2 → EmbeddingModelV3 - Update all type imports to V3 equivalents - Add comprehensive implementation documentation Known issues: - Usage structures need migration to V3 nested format - Test files need V3 type updates - Build currently fails due to usage type mismatches See AI-SDK-V6-IMPLEMENTATION.md for full details * feat: Complete AI SDK v6 support - Migrate from ProviderV2 to ProviderV3 - Migrate from LanguageModelV2 to LanguageModelV3 - Migrate from EmbeddingModelV2 to EmbeddingModelV3 - Update specificationVersion to 'v3' - Update usage structure to nested v6 format: - inputTokens: { total, noCache, cacheRead, cacheWrite } - outputTokens: { total, text, reasoning } - Update all type imports (SharedV2* -> SharedV3*, etc.) - Update dependencies: - @ai-sdk/provider: 2.0.0 -> 3.0.0-beta.27 - @ai-sdk/provider-utils: 3.0.1 -> 4.0.0-beta.53 - ai peer dep: ^6.0.0 Tested with free models on OpenRouter: - Basic text generation ✅ - Streaming ✅ - Tool calling ✅ - Reasoning tokens ✅ - Multi-turn conversations ✅ * fix: Complete AI SDK v6 type migration - Remove documentation files from PR - Fix all V2→V3 type migrations in source files - Add MSW-based test server utility to replace removed createTestServer - Migrate all test files to V3 types - Fix JSONSchema7 type issues in tests - Fix requestBody unknown type issues - Typecheck now passes with 0 errors - Build succeeds - Live tests pass (basic, streaming, tool calling, reasoning, multi-turn) * chore: update to stable AI SDK v6 releases - Update @ai-sdk/provider from 3.0.0-beta.27 to 3.0.0 - Update @ai-sdk/provider-utils from 4.0.0-beta.53 to 4.0.1 - Update ai from 6.0.0-beta.159 to 6.0.3 - Update tests for v6 usage format (nested inputTokens/outputTokens) - Update tests for v3 embedding specification version - Fix header tests for extended user-agent format * fix: Return LanguageModelV3FinishReason as object { unified, raw } In AI SDK v6, the LanguageModelV3FinishReason type changed from a string to an object with `unified` and `raw` properties. This change ensures the provider returns the correct format: - `unified`: The normalized finish reason (stop, length, tool-calls, etc.) - `raw`: The original finish reason string from the API Changes: - Update mapOpenRouterFinishReason to return { unified, raw } object - Add createFinishReason helper for creating finish reasons - Update chat and completion streaming to use new format - Update all tests to expect object format * fix: address review feedback and merge main branch fixes - Bump version to 2.0.0 - Fix zod peer dependency (^3.25.0 || ^4.0.0) - Make file content part fields optional and add file_id - Handle LanguageModelV3ToolApprovalResponsePart in tool responses - Apply biome formatting fixes * chore: pin devDependency versions * chore: pin ai peer dep * docs: update README for AI SDK v6 release --------- Co-authored-by: Robert Yeakel <212159665+robert-j-y@users.noreply.github.com>

This commit addresses issue #186 where `bun install` shows "warn: incorrect peer dependency ai@6.0.86" warning. Root cause: @openrouter/ai-sdk-provider@1.5.4 requires ai@^5.0.0 as peer dependency, but @link-assistant/agent uses ai@^6.0.1. Solution: - Update @openrouter/ai-sdk-provider from ^1.5.4 to ^2.2.3 (version 2.0.0+ supports AI SDK v6) - Update @opentui/core from ^0.1.46 to ^0.1.79 - Update @opentui/solid from ^0.1.46 to ^0.1.79 Note: The solid-js peer dependency warning remains due to upstream issue in @opentui/solid which uses exact version pinning (1.9.9). Reported at: anomalyco/opentui#689 Added case study documentation in docs/case-studies/issue-186/ References: - OpenRouterTeam/ai-sdk-provider#307 - https://github.com/OpenRouterTeam/ai-sdk-provider/releases/tag/2.0.0 Fixes #186 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

pablof7z added 2 commits December 18, 2025 01:05

AviVahl reviewed Dec 23, 2025

View reviewed changes

src/tests/usage-accounting.test.ts Outdated Show resolved Hide resolved

AviVahl mentioned this pull request Dec 23, 2025

Support AI SDK 6 (Stable) #316

Closed

rorhug mentioned this pull request Dec 26, 2025

Support for AI SDK v6 get-convex/agent#202

Closed

saeta-eth reviewed Dec 26, 2025

View reviewed changes

src/test-utils/test-server.ts Show resolved Hide resolved

The-Best-Codes reviewed Dec 29, 2025

View reviewed changes

src/tests/provider-options.test.ts Outdated Show resolved Hide resolved

The-Best-Codes reviewed Dec 29, 2025

View reviewed changes

lmorchard mentioned this pull request Jan 6, 2026

build(deps): bump @ai-sdk/openai-compatible from 1.0.27 to 1.0.29 mozilla/pilo#229

Closed

Crunchyman-ralph mentioned this pull request Jan 6, 2026

feat: add context7 as an mcp tool to taskmaster ai sdk calls eyaltoledano/claude-task-master#1447

Open

16 tasks

subtleGradient closed this Jan 7, 2026

subtleGradient mentioned this pull request Jan 7, 2026

fix(chat): forward model options including reasoning to API requests #321

Closed

robert-j-y reopened this Jan 19, 2026

robert-j-y assigned robert-j-y and unassigned subtleGradient Jan 19, 2026

robert-j-y self-requested a review January 19, 2026 18:37

robert-j-y reviewed Jan 19, 2026

View reviewed changes

package.json Outdated Show resolved Hide resolved

robert-j-y added 5 commits January 19, 2026 11:55

fix: address review feedback and merge main branch fixes

b616025

- Bump version to 2.0.0 - Fix zod peer dependency (^3.25.0 || ^4.0.0) - Make file content part fields optional and add file_id - Handle LanguageModelV3ToolApprovalResponsePart in tool responses - Apply biome formatting fixes

chore: pin devDependency versions

512b100

chore: pin ai peer dep

a3b73c1

Merge branch 'main' into feature/ai-sdk-v6-support

c21ce53

docs: update README for AI SDK v6 release

777c0a7

robert-j-y merged commit 6fd68db into OpenRouterTeam:main Jan 19, 2026
1 check passed

github-actions bot mentioned this pull request Jan 19, 2026

Version Packages #325

Merged

robert-j-y mentioned this pull request Jan 27, 2026

Support Provider v3 for AI SDK v6 beta #203

Closed

konard mentioned this pull request Feb 15, 2026

fix: resolve incorrect peer dependency warning for ai@6.0.86 link-assistant/agent#189

Merged

4 tasks

Conversation

pablof7z commented Dec 17, 2025

Summary

Changes

Testing

Notes

Uh oh!

socket-security bot commented Dec 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

AviVahl commented Dec 23, 2025

Uh oh!

seannetlife commented Dec 23, 2025

Uh oh!

elijah629 commented Dec 24, 2025

Uh oh!

Uh oh!

saeta-eth commented Dec 26, 2025

Uh oh!

ravisojitra commented Dec 28, 2025

Uh oh!

ZenAlexa commented Dec 28, 2025

Uh oh!

pablof7z commented Dec 28, 2025

Updated to stable AI SDK v6 releases

Uh oh!

Uh oh!

The-Best-Codes left a comment

Choose a reason for hiding this comment

Uh oh!

wong2 commented Dec 29, 2025

Uh oh!

ldriss commented Dec 29, 2025

Uh oh!

fwang2002 commented Dec 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vzt7 commented Dec 31, 2025

Uh oh!

ldriss commented Dec 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mattapperson commented Jan 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ldriss commented Jan 1, 2026

Uh oh!

pablof7z commented Jan 1, 2026

Fixed: LanguageModelV3FinishReason type

Uh oh!

ldriss commented Jan 1, 2026

Uh oh!

nvti commented Jan 4, 2026

Uh oh!

wlib commented Jan 4, 2026

Uh oh!

ldriss commented Jan 5, 2026

Uh oh!

The-Best-Codes commented Jan 5, 2026

Uh oh!

haowang1013 commented Jan 7, 2026

Uh oh!

ldriss commented Jan 7, 2026

Uh oh!

subtleGradient commented Jan 7, 2026

Uh oh!

ldriss commented Jan 7, 2026

Uh oh!

ZenAlexa commented Jan 7, 2026

Uh oh!

cpakken commented Jan 7, 2026

Uh oh!

The-Best-Codes commented Jan 7, 2026

Uh oh!

Uh oh!

Uh oh!

robert-j-y commented Jan 23, 2026

Uh oh!

Reviewers

socket-security bot commented Dec 17, 2025 •

edited

Loading

fwang2002 commented Dec 30, 2025 •

edited

Loading

ldriss commented Dec 31, 2025 •

edited

Loading

mattapperson commented Jan 1, 2026 •

edited

Loading