Release Notes - v1.25.21
API alignment: Aligned SDK method signatures with HTTP API definitions by adding explicit parameters.
Compatibility improvement: Added base_compatible_api_url support for OpenAI-compatible API scenarios.
OSS fixes: Fixed OSS URL wrapping and related CLI cleanup issues.
Async fixes: Fixed async cancel/list behavior and aiohttp SSE flattened_output handling.
Quality improvements: Fixed multiple code defects, resolved pre-commit lint issues, and removed unnecessary generated files.
Documentation: Expanded README coverage for SDK and CLI usage.

What's Changed

• Dev/sdk api alignment by @luk384090-cloud in #132

Full Changelog: v1.25.20...v1.25.21

Release date: 2026-06-01

This release centers on a complete CLI rewrite. There are no breaking changes to the runtime SDK API — Python callers do not need to change anything. CLI users should read
the "Command Changes" and "Compatibility" sections below.

✨ Highlights

CLI framework migration

• Migrated the entire CLI from argparse to Typer, bringing modern subcommand UX, shell completion, and rich help output.
• Split the monolithic dashscope/cli.py (843 lines) into a modular package dashscope/cli/ with dedicated submodules: generation, fine_tunes, files, deployments, oss, and
  agentic_rl.
• Added a python -m dashscope entry point (dashscope/cli/main.py).

New commands

• dashscope rl — Agentic-RL fine-tuning command group, loaded lazily so missing optional dependencies will not break other commands.
• dashscope ft — short alias for fine-tunes (recommended going forward). fine-tunes still works but is now hidden from help.

What's Changed

• 开发/命令行重构 by @luk384090-cloud in #128
• Dev/cli refactor by @luk384090-cloud in #127
• Dev/CLI Refactor by @luk384090-cloud in #129
• Dev/CLI Refactor by @luk384090-cloud in #130

New Contributors

• @luk384090-cloud made their first contribution in #128

Full Changelog: v1.25.19...v1.25.20

Highlights

This release introduces Agentic Reinforcement Learning (RL) — a complete SDK and CLI for managing RL fine-tuning workflows on the DashScope platform, including function registration, dataset management, job lifecycle, and observability.

New Features

Agentic RL SDK (AgenticRL)

• Full workflow support: function registration, dataset upload, job submission, status query, cancellation, and deletion
• YAML-driven configuration for reproducible training workflows
• Path-driven mode with automatic function packaging and upload
• Support for Rollout, Reward, and Group Reward function components
• Async-first API design with asyncio support

Agentic RL CLI (dashscope rl)

• Production-grade CLI built with Typer and Rich
• Commands: register_functions, test_functions, upload_data, run, get, list, cancel, logs
• Multiple output formats: table, json, yaml
• --verbose flag controls log verbosity (default: WARNING) and enables root-cause-only tracebacks

ReadME

What's Changed

• Add headers to the response. by @mose-zm in #126
• sdk/cli: agentic rl by @lzsweb in #125

New Contributors

• @lzsweb made their first contribution in #125

Full Changelog: v1.25.18...v1.25.19

What's Changed

• (model/cosyvoice): support flush api by @songguocola in #124

Full Changelog: v1.25.17...v1.25.18

What's Changed

• add http sse cosyvocie tts api and omni createItem function by @songguocola in #122

Full Changelog: v1.25.16...v1.25.17

Full Changelog: v1.25.15...v1.25.16

Full Changelog: v1.25.14...v1.25.15

What's Changed

• support wan27 by @mose-zm in #117

Full Changelog: v1.25.13...v1.25.14

What's Changed

• feat(model/cosyvoice): add custom input by kwargs in create_voice by @songguocola in #113

Full Changelog: v1.25.12...v1.25.13

What's Changed

• feat(model/cosyvoice): add hot_fix and max_prompt_audio_length param by @songguocola in #110
• feat(model/cosyvoice): fix hot_fix setting by @songguocola in #111

Full Changelog: v1.25.11...v1.25.12