Skip to content

CHANGELOG.md

Latest commit

 

History

History
73 lines (46 loc) · 6.73 KB

CHANGELOG.md

File metadata and controls

73 lines (46 loc) · 6.73 KB
Raw

Changelog

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

1.26.0 - 2026-07-08

Added

  • GenerateOptions: Added thinking_budget, logprobs, top_logprobs, search_options, parallel_tool_calls, response_format, output_format parameters for thinking budget control, log probabilities, web search configuration, parallel tool calls, JSON mode, and deep-research output format.
  • MultiModalConversationOptions: Added ocr_options, logprobs, top_logprobs parameters for OCR model support and log probabilities.
  • ImageSynthesisOptions: Added explicit seed, style, ref_strength, ref_mode, prompt_extend, watermark, bbox_list, enable_sequential, thinking_mode, color_palette parameters (previously only via kwargs).
  • VideoSynthesisOptions: Added shot_type and audio_setting parameters for video shot type and audio configuration.

Fixed

  • CodeGeneration: Removed services/ prefix from API path (aigc/code-generation/generation), aligning with Python is_service=False fix in v1.26.
  • Understanding: Removed services/ prefix from API path (nlp/nlu/understanding), aligning with Python is_service=False fix in v1.26.
  • Models.get(): Rewritten to use query-parameter filtering (?model={name}&page_no=1&page_size=1) instead of path-based lookup, returning 404 when model is not found. Aligned with Python Models.get() rewrite in v1.26.
  • WebSocket duplex: Added readyState check before ws.send() in runWebSocketDuplexTask to prevent errors when the connection closes mid-stream. Aligned with Python WebSocket connection-check fix in v1.26.
  • UTF-8 encoding: Updated Accept and Content-Type headers to include charset=utf-8, ensuring non-ASCII characters (e.g. Chinese) are transmitted correctly. Aligned with Python ensure_ascii=False fix in v1.26.

Changed

  • Models.get() (Breaking): Return shape changed from raw response body to { ...response, output: model } on success, or { status_code: 404, message, output: null } when the model is not found. Previously returned the raw response body from path-based lookup (GET /models/{name}). Callers that accessed result.output directly will still work; callers that accessed other top-level fields should use the new spread-based return shape.

Synced from dashscope-sdk-python v1.26.2 (tag v1.26.2).

1.25.5 - 2026-07-08

Added

  • ImageGeneration (src/aigc/imageGeneration/): New module for wan2.6-image / wan2.6-t2i image generation, based on a messages interface. Supports synchronous (streaming / non-streaming) and asynchronous task modes. Includes call(), asyncCall(), fetch(), wait() methods, with incremental_to_full streaming merge and wait_timeout support. Exported as ImageGeneration and ImageGenerationModels from the package entry.
  • Async task wait_timeout: Added optional wait_timeout parameter (in seconds) to ImageSynthesis, VideoSynthesis, Transcription, and BatchTextEmbedding call() methods. When set to a value > 0, the method returns a timeout response (code: 'WaitTaskTimeout', status_code: 408) instead of waiting indefinitely. Aligned with Python BaseAsyncApi.wait(wait_timeout).
  • MultiModalConversation incremental merge: Added incremental_to_full streaming logic to MultiModalConversation — when incremental_output is false and the model supports it, the SDK transparently requests incremental output and merges deltas into a full response, matching Generation behavior.

Fixed

  • Generation User-Agent header: The incremental_to_full User-Agent flag now includes the SDK version (dashscope-sdk-nodejs/x.y.z; incremental_to_full/N) instead of overriding the entire User-Agent string. Previously the SDK version was lost during streaming requests.

Synced from dashscope-sdk-python v1.25.24 (tag v1.25.24).

1.25.4 - 2026-06-09

Added

  • GenerateOptions: Added explicit presence_penalty, tools, tool_choice, enable_thinking, n parameters for function calling and reasoning mode support.
  • MultiModalConversationOptions: Added explicit temperature, top_p, top_k, max_tokens, seed, stop, repetition_penalty, presence_penalty, result_format, incremental_output, enable_search, tools, tool_choice, enable_thinking, n parameters.
  • VideoSynthesisOptions: Added explicit seed, prompt_extend, watermark, resolution, ratio parameters.
  • TextEmbeddingOptions: Added dimension, output_type, instruct parameters for v3/v4 models.
  • MultiModalEmbeddingOptions: Added dimension, output_type, fps, instruct, enable_fusion, res_level, max_video_frames parameters for qwen3-vl-embedding and snapshot models.
  • TextReRankOptions: Added instruct parameter for custom ranking instructions.

Synced from dashscope-sdk-python v1.25.21 (tag v1.25.21).

1.25.3 - 2026-04-24

Added

  • HTTP TTS (HttpSpeechSynthesizer): New HTTP-based text-to-speech interface for CosyVoice, supporting both streaming (SSE) and non-streaming modes. Synced from dashscope-sdk-python v1.25.17 (tag v1.25.17).
    • Non-streaming: Returns audio URL, audio ID, and expiration timestamp.
    • Streaming: Yields audio data chunks with sentence-level timestamps.
    • Added HttpSpeechSynthesisResult class with getters for audioData, audioUrl, audioId, expiresAt, sentences, and response.

1.25.2 - 2026-04-07

Re-export result types and classes from the package entry for migration from legacy deep imports and typing: BaseResult, GenerationResult, EmbeddingResult, MultiModalEmbeddingResult, AsyncTaskResult, RecognitionResult.

1.25.1 - 2026-04-03

First open-source release on GitHub and npm as dashscope-sdk-official, migrated from the internal @ali/dashscope-sdk package. Aligned with dashscope-sdk-python v1.25.15 (tag v1.25.15).