v2.8.1

v2.8.1 is a provider-compatibility patch release. It updates the bundled DeepSeek and Qwen integrations to their current OpenAI-compatible API contracts, adds provider examples and live smoke tests, and makes documentation release verification less dependent on third-party cache behaviour.

Provider Updates

DeepSeek V4

The DeepSeek presets and evaluation connections now use the current V4 model identifiers: deepseek-v4-flash and deepseek-v4-pro. The capability matrix no longer infers behaviour from the legacy reasoner model-name substring. V4 requests forward thinking and reasoning_effort, support tool calling and JSON Output, and degrade native JSON Schema requests to the provider-supported JSON object format.

Reasoning-content coverage now includes an opt-in DeepSeek V4 live smoke test.

Qwen

The Qwen preset now targets qwen3.8-max through DashScope's international OpenAI-compatible endpoint. It recognizes QWEN_API_KEY, current context/output limits, Qwen thinking controls, and model-specific JSON Schema support. Unsupported tool-choice modes are translated to valid Qwen requests instead of being sent unchanged.

New structured-output and LLM API examples are included under examples/, with deterministic request/capability coverage and an opt-in live smoke test.

Documentation and Release Verification

The documentation release gate now verifies the uniquely versioned release page on both published documentation sites and relies on the generated release-note parity check for navigation coverage. This avoids treating a stale third-party CDN copy of the long-lived release index as a release failure.

Upgrade Notes

The Qwen bundled preset now reads QWEN_API_KEY; users who previously configured only DASHSCOPE_API_KEY should set the new variable or provide an explicit Qwen configuration.

The deepseek and deepseek-r preset names remain available, but now select DeepSeek V4 Flash and V4 Pro respectively. No other breaking changes are intended.