Internal

• Improve CLI usage statistics.

Added

• Linux ARM64 binaries: Add Linux ARM64 packages (glibc), supporting standalone installation and binary updates.

Fixed

• Image results: Preserve all images returned by concurrent asynchronous image generation and editing tasks, including tasks with different response formats.
• Streaming chat: Detect prematurely closed Chat Completions streams, preserve complete tool_calls, and return finish_reason in structured output.
• CLI shutdown: Prevent pending telemetry requests from delaying exit and preserve queued events when telemetry initialization fails.

Added

• Aliyun CLI integration: Support invocation through aliyun bl, with host-aware help, examples, recovery hints, and structured commands.
• AK/SK authentication: Support BAILIAN_ACCESS_KEY_ID, BAILIAN_ACCESS_KEY_SECRET, and optional BAILIAN_SECURITY_TOKEN. Automatically obtain Console tokens with AK/SK when missing or expired.
• Custom model import: Add bl cmodel base-models/import/get/list/watch/delete to import custom models from OSS, list supported base models, track import jobs, and query or delete imported models. The import API is currently available in the Beijing region only.

Changed

• Command output: JSON mode returns complete business results, with warnings, diagnostics, and errors written separately to stderr as JSONL for scripts and Agents.
• Quiet mode: Hide only progress and informational messages while preserving complete results, warnings, and errors. Support DASHSCOPE_QUIET=1. Scripts extracting IDs, paths, or answers through quiet mode should explicitly read fields from JSON output.
• Credential precedence: Resolve AK/SK in this order: command parameters, Bailian environment variables, Bailian configuration, then Alibaba Cloud environment variables.
• Model recommendations: Disable thinking for recommendation LLM calls; keep --no-thinking accepted for compatibility.
• Output mode validation: Explicitly validate supported formats for interactive login, local UI, Pipeline event streams, and streaming speech synthesis to avoid mixed output. Use text output for native interactive modes and Pipeline JSONL events, and specify --out for streaming speech synthesis in JSON mode.

Fixed

• Speech recognition: Preserve complete responses and transcriptions in JSON output, including task information when no transcriptions are returned.
• Managed Agent warnings: Preserve SDK warnings in quiet and JSON modes.
• Update notices: Display available updates even when using cached version results.
• Error diagnostics: Avoid duplicate error messages and improve request diagnostics for model recommendations and fine-tuning fee queries.

Beta build (npm dist-tag web-search). CDN rolling pointer: sync-release.json.

Added

• AIGC auto mode: bl image generate / bl image edit / bl video generate / bl speech synthesize / bl speech recognize add --model auto, routing AIGC models automatically via Bailian's self-developed MCP.

Changed

• MCP client migration: Replace the hand-rolled SSE MCP client with the official MCP SDK.

Fixed

• Token Plan quota queries: Display all quota windows returned by the API, and add hasQuotaData to JSON output to distinguish missing quota data from a normal result.

Added

• ParseX parsing: Add bl parsex parse submit/result for document and audio/video parsing, result queries, and Markdown output.
• ParseX field extraction: Add bl parsex extract submit/result to extract fields with a JSON Schema from documents or existing parsing results, including source citations.
• Configuration source diagnostics: Show API key and Base URL sources in bl config show, and warn when requests use values from different configuration sources.

Fixed

• Workspace API key login: Fix bl auth login rejecting valid workspace API keys on MaaS endpoints.
• Login validation errors: Preserve service and network errors during API key validation instead of reporting every failure as an invalid key.

Changed

• Default telemetry output (BREAKING): Business results from bl alert, bl monitor, and bl log now default to JSON, even when the Profile specifies text. Scripts relying on text must pass --output text or set DASHSCOPE_OUTPUT=text.

Fixed

• Workspace identity: Resolve the selected workspace's Console identity automatically and share it across preflight, business requests, asynchronous polling, and follow-up requests, fixing ineffective --workspace-id filtering and incorrect cross-workspace results without changing saved configuration.
• Audit log parsing: Normalize timestamps, models, request IDs, workspace IDs, and usage; fix missing fields, NaN timestamps, and malformed count displays, and reject audit results from a mismatched workspace.
• Log counts: Use each corresponding list API's totalCount for log audit count and log inference count, fixing zero counts for non-empty lists without paginating or fetching full log content.
• Inference count filters and completion: Preserve multiple-model and multiple-API-key filters by deduplicating and querying single-value combinations. Support asynchronous polling and emit totals only after every query succeeds; reject empty filters, invalid counts, and aggregate overflow.
• Dry-run previews: Show the workspace identity resolution plan and actual count request combinations without sending business requests.

Beta build (npm dist-tag mcp). CDN rolling pointer: sync-release.json.

What's Changed

• fix(files): use the active endpoint for temporary uploads by @Uddoo in #271
• fix: resolve Node 26 fetch and dispatcher incompatibility by @XXPermanentXX in #281
• Feat/batch command by @gujieye in #282

New Contributors

• @Uddoo made their first contribution in #271

Full Changelog: v2.6.0...v2.6.1