ai@7.0.146
Vercel AI SDK 7.0.146 adds maxOutputTokens and reasoning options to experimental decision calls, while resolving type errors for optional tools and cleaning up orphaned reasoning when pruning tool calls.
Why this signal matters
The ai@7.0.146 patch release introduces several updates to tool calling and experimental decisions. Specifically, maxOutputTokens and reasoning options are now supported in experimental decision calls and forwarded to language-model-backed decisions (language-model adapters default reasoning to disabled when omitted, and native providers warn if unsupported). In addition, orphaned reasoning parts (including consecutive summary and reasoning-file parts) are now properly removed when pruning tool calls. Finally, a typing fix ensures optional tools do not inject an erroneous undefined member into tool call, result, error, denied output, or execution event types. Corresponding updates were applied to @ai-sdk/gateway, @ai-sdk/provider, and @ai-sdk/provider-utils.
Actionable summary
ai@7.0.146 forwards maxOutputTokens and reasoning to language-model-backed decisions, drops orphaned reasoning parts during tool pruning, and removes incorrect undefined members from optional tool event types.
- Agent usefulness
- 77/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade the ai package to 7.0.146 along with related @ai-sdk dependencies if using experimental decision calls or optional tool types.