Alibaba Cloud Model Studio:Changelog

Version history of the Bailian CLI

2.8.0 | 2026-11-03

• The CLI can be invoked through aliyun bl, with help, examples, recovery hints, and structured commands adapted to that entry point.
• Added AK/SK authentication with BAILIAN_ACCESS_KEY_ID, BAILIAN_ACCESS_KEY_SECRET, and optional BAILIAN_SECURITY_TOKEN. When a console token is missing or expired, the CLI can obtain one automatically with AK/SK.
• JSON mode returns complete business results; warnings, diagnostics, and errors are written separately to stderr as JSONL.
• Quiet mode hides only progress and informational messages while preserving complete results, warnings, and errors; DASHSCOPE_QUIET=1 is supported. Scripts that extracted IDs, paths, or answers from quiet output must read JSON fields instead.
• AK/SK resolution order is command parameters, Bailian environment variables, Bailian configuration, then Alibaba Cloud environment variables.
• Model recommendations disable thinking; --no-thinking remains accepted for compatibility.
• Interactive login, local UI, Pipeline event streams, and streaming speech output validate their supported formats: native interactive modes and Pipeline JSONL events use text output, and streaming speech JSON output requires --out.
• Fixed speech recognition JSON output dropping the complete response and transcription; empty transcriptions still return task information.
• Available updates are still shown when cached version results are used.
• Avoided duplicate error messages and improved request diagnostics for model recommendations and fine-tuning fee queries.

2.7.2 | 2026-10-30

• bl image generate, bl image edit, bl video generate, bl speech synthesize, and bl speech recognize add --model auto, which routes AIGC models automatically through Bailian's self-developed MCP.
• Replaced the hand-rolled SSE MCP client with the official MCP SDK.

2.7.1 | 2026-10-29

• bl usage token-plan now displays every quota window returned by the API; JSON output adds hasQuotaData to distinguish missing quota data from a normal result.

2.7.0 | 2026-10-29

• bl config show now shows the effective sources of the API Key and Base URL and warns when they come from different configuration sources.
• Fixed bl auth login rejecting valid workspace API keys on MaaS endpoints.
• API Key validation now preserves service and network errors instead of reporting every failure as an invalid key.

2.6.1 | 2026-10-28

• Added bl batch convert/create/list/get/cancel/download for CSV-to-JSONL conversion, Chat or Embedding task creation, progress tracking, cancellation, and result/error downloads with optional CSV export.
• bl text chat --batch sends one non-streaming request, separate from asynchronous file-based batch tasks, with request timeouts of 60–3600 seconds. It requires a Batch-capable base URL and rejects international, realtime, and Token Plan endpoints.
• bl advisor recommend adds --top, --model, --no-agent, --trace, --think, and --no-thinking.
• Model recommendations upgrade intent analysis, routing, candidate recall, scoring, and ranking. Interactive terminals support Agent clarification and multi-step recommendations while preserving the existing recommendation JSON envelope.
• Temporary file uploads now select the upload service from the active model endpoint, fixing Singapore and Prime video uploads incorrectly using Beijing. Official temporary uploads support only Beijing and Singapore; unsupported regions and Token Plan fail explicitly.
• Removed legacy recommendation library interfaces (BREAKING), including getModels, recallCandidates, recallSemantic, and rankModels. Library consumers must adapt to the new recommendation engine; legacy types are exported as LegacyIntentProfile, LegacyModelProfile, and LegacyRecommendResult.
• Fixed fetch/dispatcher incompatibility and proxy initialization on Node.js 26.
• Nested business errors are recognized correctly, and non-JSON error bodies and request IDs are preserved instead of treating error responses as successful results.

2.6.0 | 2026-10-27

• Added --introspect to print the machine-readable command schema (JSON), so Agents can discover commands and parameters.

2.5.1 | 2026-10-23

• Console login supports --no-browser to print the login URL manually and --token-stdin to import and validate console credentials from standard input.
• Commands can declare the calling Agent, Skill name, and Skill version; undeclared callers are recorded as unknown.
• Fixed login completion when the browser callback returns quickly, and report a timeout when no credentials are received.

2.3.3 | 2026-10-19

• Fixed model API requests failing in the binary CLI, including speech and video generation.
• Fixed long-running API requests timing out before the configured --timeout.
• Fixed console login URLs losing query parameters when opening the browser on Windows.
• After console login or when running bl auth status, clearly indicate when a model API key is missing.

2.3.2 | 2026-10-16

• bl update now exits with a non-zero status when an upgrade fails or the latest version cannot be determined.

2.3.1 | 2026-10-16

• Requests configured with a timeout longer than 300 seconds are no longer interrupted by the default response-header timeout.

2.3.0 | 2026-10-14

• Added Pi support: bl config agent accepts --agent pi.
• Global flags can precede the full command path, for example bl --config token-plan auth status.
• Unknown flags, missing values, and invalid values return usage exit code 2.
• --timeout now sets the transfer timeout when uploading local files to temporary OSS storage.

2.2.2 | 2026-10-14

• bl config agent now supports Pi, writing its native configuration while preserving existing settings.
• Global flags can precede or follow the command; both forms select the Profile for that invocation only.
• Only flags defined for a command are accepted; unknown flags, missing values, and invalid values return usage exit code 2.
• Local file uploads to temporary OSS storage honor --timeout instead of a fixed 120-second limit.

2.2.1 | 2026-10-12

• bl config agent --agent hermes now writes to the location Hermes actually reads: $HERMES_HOME/config.yaml, or the platform-native default.

2.2.0 | 2026-10-12

• bl usage token-plan reports a single monthly quota (usage percentage and reset time) instead of the previous 5-hour and 1-week windows.

2.1.0 | 2026-09-28

Fixed Skill installation on Windows: legacy path length limits are handled, and installed Skills update in place when running agents keep directories open.

2.0.1 | 2026-09-22

bl text chat and bl video generate add --prime to call through a workspace-scoped endpoint in Prime mode, with --workspace-id to specify the workspace.

2.0.0 | 2026-09-22

Skill installs and updates extend to WorkBuddy, Trae CLI, and DeepSeek DSH; bl skill init, bl skill add, bl skill update, and bl skill remove support --dry-run previews.

1.27.0 | 2026-09-18

New bl model search to search the model catalog by relevance, and bl model code to get SDK example code.

1.26.0 | 2026-09-17

Simplified authentication: console login is the recommended default and can create a standard API key automatically when needed; API keys are validated before saving, and the CLI selects an available region automatically. Existing standard API keys and Token Plan subscription keys now log in uniformly with bl auth login --api-key.

1.23.0 | 2026-09-10

ASR instant hot words (--vocabulary) and contextual enhancement (--context); speech vocabulary management commands.

1.18.0 | 2026-08-27

Profile API key auto-fallback; default video model upgraded to Wan3.0.

1.17.0 | 2026-08-18

CLI available in Chinese and English.

1.14.0 | 2026-08-04

Node.js-free binary install and version-pinned updates.

For issues or ideas, open one on GitHub. See the full record of Bailian CLI feature updates, improvements, and fixes in CHANGELOG.md.