v0.2.3 - Fix xAI Grok reasoningEffort Error

What's Changed

Bug Fix

Fixed xAI Grok models failing with reasoningEffort parameter error

xAI Grok models use model variant selection (grok-4-1-fast-reasoning vs grok-4-1-fast-non-reasoning) to control reasoning mode, not the reasoningEffort API parameter. The provider was incorrectly sending reasoningEffort for xAI reasoning models, causing:

HTTP 400: Model does not support parameter reasoningEffort

Note on Model IDs

The correct xAI Grok 4.1 model IDs use hyphens (not dots):

• xai.grok-4-1-fast-reasoning
• xai.grok-4-1-fast-non-reasoning

If you have xai.grok-4.1-fast-reasoning (with dots) in your config, update to use hyphens.

Full Changelog: v0.2.2...v0.2.3

v0.2.2 - Fix Gemini Streaming Tool Calls

What's Changed

Bug Fix

Fixed Gemini streaming tool calls not working - Tool calls were being ignored in streaming mode because OCI's Generic format returns them at message.toolCalls rather than inside message.content[].

This fix enables proper tool calling in OpenCode when using Gemini models with streaming (the default mode).

Full Changelog: v0.2.1...v0.2.2

v0.2.1 - Gemini Parallel Tool Calls Fix

What's Changed

Bug Fixes

• Gemini parallel tool calls: Fixed issue where Gemini models failed when multiple tool calls occurred in a single turn. OCI's Generic format doesn't properly support parallel function calls for Gemini, so we now use a text fallback approach (same as Llama/xAI) when 2+ tool calls are detected.

• Finish reason detection: Fixed finishReason to correctly return tool-calls when tool calls are present in the response.

• Zod v4 schema support: Added ensureJsonSchema() helper that converts Zod v4 schemas to JSON Schema when needed (AI SDK v5 compatibility).

Other Changes

• Updated .gitignore to exclude local test files and tool directories
• Bumped ai SDK to 5.0.126
• Added zod-to-json-schema dev dependency

Technical Details

The parallel tool call fix works by:

• Single tool call: Uses standard toolCalls array + TOOL message (works correctly)
• Parallel tool calls (2+): Converts both tool calls and results to text format in ASSISTANT/USER messages

This is a workaround for OCI's Generic format limitation that doesn't properly batch multiple TOOL responses for Gemini.

Full Changelog: v0.2.0...v0.2.1