# What changed in @ai-sdk/gateway

> 54 changes in @ai-sdk/gateway 2 to 4 from its official release notes, upgrade guides and advisories: 4 breaking, 26 new APIs and options. Each one cited. Sources checked 2026-10-18. HTML: https://unstale.sh/npm/@ai-sdk/gateway/

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

## @ai-sdk/gateway 4

### Breaking changes

- Arrays passed directly as state now contain decision state parts; wrap JSON arrays in an object or a json part. (4.0.108) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.108)
- The API name evaluate was renamed to decide. (4.0.104) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.104)
- Unsupported batch request types are now rejected. (4.0.79) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.79)
- Support for Imagen models and their integration was dropped following the API shutdown. (4.0.59) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.59)

### New APIs and options

- Adds support for the GPT-6.1 Sol model. (2.0.161, 3.0.206, 4.0.101) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.101), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.206), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.161)
- GatewayProviderOptions now accepts conditional evaluation fallbacks. (4.0.96) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.96)
- Gemini 3.8 speech model IDs were added to Gateway types. (4.0.91) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.91)
- Adds model IDs for GPT-6 Sol and Luna. (2.0.156, 3.0.199, 4.0.90) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.90), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.199), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.156)
- Adds support for cancelling batches with experimental_cancelBatch. (4.0.88) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.88)
- Adds the Grok 4.7 model ID. (2.0.154, 4.0.88) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.88), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.154)
- The has model filter now supports reasoning. (4.0.88) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.88)
- The has model filter now supports tool-use. (4.0.88) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.88)
- Image generation requests are now supported in batches. (4.0.79) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.79)
- Batch requests can now use per-request models. (4.0.76) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.76)
- Adds support for the gpt-6-astra model in the gateway. (2.0.147, 3.0.189, 4.0.75) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.75), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.189), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.147)
- Adds support for Fable 5.1, allowing developers to use it through the Gateway. (2.0.145, 3.0.187, 4.0.72) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.72), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.187), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.145)
- Accepts a webhookUrl for batch completion webhooks; the gateway provider registers it through the batch callbackUrl contract. (4.0.64) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.64)
- Adds Tako Search tool support to the gateway provider. (3.0.180, 4.0.62) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.180), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.62)
- DeepSeek V4 Flash Vision Exp now supports image input. (2.0.138, 3.0.178, 4.0.60) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.60), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.178), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.138)
- DeepSeek V4 Flash Vision Exp now supports the Files API. (4.0.60) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.60)
- Adds experimental batch API support through the AI Gateway batch routes, with methods to start text batches and check their status and results. (4.0.57) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.57)

### Behavior and default changes

- The gateway sends image files in decision state. (4.0.110) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.110)
- Decision refusals are reported as refusal answers. (4.0.106) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.106)
- Conditional evaluation fallback models with any non-empty value are now accepted. (4.0.96) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.96)
- The has provider option now supports quantization conditions. (4.0.91) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.91)
- The video model now implements handleWebhookOption, so generateVideo({ webhook }) registers the factory URL as callbackUrl and awaits delivery instead of falling back to polling. (4.0.60) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.60)

### Tooling

- Conditional evaluation fallbacks are now typed in the Gateway provider. (4.0.95) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.95)

### Fixes and other changes

- Batch result downloads limit JSONL rows to 64 MB. (4.0.105) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.105)
- getCredits() now forwards the configured team as a query parameter, preventing authentication errors for credentials that can access multiple teams. (4.0.102) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.102)
- getAvailableModels() now includes evaluation models instead of filtering them out as unknown model types. (4.0.85) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.85)
- The gateway provider now forwards server-returned warnings in doGenerate instead of overwriting them with an empty array. (2.0.151, 3.0.193, 4.0.80) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.80), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.193), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.151)
- Preserves retryability returned by the AI Gateway for image results. (4.0.77) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.77)
- Transient network errors while reading successful response bodies are now marked retryable, including for AI Gateway responses. (2.0.144, 3.0.185, 4.0.68) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.144), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.185), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%404.0.68)

## @ai-sdk/gateway 3

### New APIs and options

- Adds a provider option for resuming responses that used up their maxOutputTokens budget. (3.0.213) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.213)
- Adds the gemini-4-argon model id. (2.0.166, 3.0.213) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.213), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.166)
- Exposes the first request's prompt token count in Google provider metadata. (2.0.166, 3.0.213) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.213), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.166)
- Adds Browserbase Search and Fetch tool support. (3.0.202) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.202)
- The has provider option now supports quantization conditions. (2.0.157, 3.0.200) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.200), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.157)
- Model filtering with has now supports reasoning and tool-use. (2.0.154, 3.0.197) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.197), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.154)
- Added the gemini-3.8-flash model for use with the Google provider. (2.0.146, 3.0.188) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.188), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.146)
- Adds support for the GLM-5.3-Flash model through the Z.AI provider and AI Gateway. (3.0.183) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.183)

### Behavior and default changes

- generateText with the Interactions API now polls background model calls until they finish. (3.0.213) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.213)
- Requests that reach the per-request CONTINUATION cap continue automatically, with usage summed across requests. (3.0.213) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.213)
- The has provider option now accepts structured-output. (2.0.159, 3.0.202) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.202), [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.159)
- Retryability classification is provider-independent and preserved through the AI Gateway; Google and Google Vertex content-filtered results are terminal. (3.0.191) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.191)

### Fixes and other changes

- Unclassified empty image results are retried, and retry-attempt accounting is preserved. (3.0.191) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%403.0.191)

## @ai-sdk/gateway 2

### New APIs and options

- Adds a provider option and provider metadata for resuming responses that used up their maxOutputTokens budget. (2.0.166) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.166)

### Behavior and default changes

- Requests that reach the per-request CONTINUATION cap are continued automatically, with usage summed across requests and the first request's prompt token count exposed. (2.0.166) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.166)
- Thinking can be disabled up to high effort; when disabled, xhigh and max effort are lowered to high with a warning. (2.0.164) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.164)
- Thinking token budgets are unsupported; enabled thinking with budgetTokens is converted to adaptive thinking with a warning. (2.0.164) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.164)
- Adds provider-independent retryability classification and preserves it through the AI Gateway. (2.0.149) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.149)

### Fixes and other changes

- Diagnostics from completed attempts are preserved, making them available after retries. (2.0.149) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.149)
- Unclassified empty image results are retried, changing how image-generation failures are handled. (2.0.149) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.149)
- Google Vertex RAI-filtered results are marked terminal, so they are not treated as retryable. (2.0.149) — [release notes](https://github.com/vercel/ai/releases/tag/%40ai-sdk/gateway%402.0.149)