Masthead 1.0.1

Masthead's AI features now work with real AI provider plugins. Before 1.0.1 they had only run against a test double. With AI Provider for Anthropic, every AI request either went to an older model or failed. Upgrade if you use Masthead's AI.

masthead-demo.mp4

Fixed

• Anthropic models. Masthead always sent a temperature, but Claude 5 models reject sampling parameters. The WordPress AI Client therefore skipped the configured claude-haiku-5-5 and claude-sonnet-5-5 and picked an older model, or none. No temperature is sent now; a site can still set one through the masthead_config filter.
• JSON schemas. Anthropic's structured outputs reject minLength, minimum and maxItems with a 400 error. Masthead now sends the provider a schema without them, states each one in the field's description instead, and still checks every reply against the full schema.
• Provider check. "Provider: available" used to mean the provider could generate any text at all. It now means a model can take a system instruction and return JSON that matches a schema, which every Masthead request needs. A provider plugin that can't do that, such as OpenRouter 1.0.0, shows as Provider: none Masthead can use instead of failing on every request.

Which AI provider plugins work

Plugin	Status
AI Provider for Anthropic 1.0.5	✅ Run end to end against the real API: ideas, outlines, share copy and series proposals
AI Provider for OpenAI 1.2.0, Google 1.2.1, Mistral 1.3.1, Ollama 1.4.0	✅ Declare schema output (set your own model ids)
AI Provider for OpenRouter 1.0.0, DeepSeek 1.0.3	❌ No schema output yet

The README has the details, and it now plays the demo video inline.

Full details are in CHANGELOG.md.