Providers and evidence

The Tier-1 portable contract covers OpenAI, Anthropic, Azure OpenAI, Gemini, Vertex, Qwen, Kimi/Moonshot, DeepSeek, Meta Standard and a configured vLLM deployment.

from zhivex_ai import create_openai, create_anthropic, create_qwen, create_meta

provider(model_id) uses the SDK-owned portable interface. Provider-native resources and hosted tools have narrower compatibility guarantees. Meta Standard and Contributor are separate targets; a hosted model on another API is not a direct integration with the model’s author.

Contract tests, installed-wheel checks and authenticated live certification are separate evidence levels. Certification is specific to the artifact, model, operation set and time window. This site does not certify a provider.

Consult the provider guide and the release’s support policy. Their links are pinned to the documented artifact’s source revision.

September provider updates

TypeScript 1.21.1 and Python 0.25.0 include Qwen3.8-Omni-Flash support. The TypeScript adapter is @zhivex-ai/qwen 0.12.0. The model accepts text, images, audio, and video and returns text over HTTP. Streaming, callable tools, and hosted web search through Responses are supported. Use prompted structured output with local validation; native JSON Schema and audio output are not supported.

DeepSeek V4.1 Flash adds image input. Gemini 3.8 Live adds catalog and protocol updates, including background thinking. OpenAI managed sessions and Anthropic compaction use native Beta APIs; Python 0.26 promotes the documented GPT-Live WebSocket subset to Stable; native WebRTC remains Experimental. Model catalog entries are distinct from artifact-specific live certification.

Published live contracts

Normalized live-agent and realtime contracts are Stable through zhivex_ai.live. Use the documented GPT-Live WebSocket subset for durable client delegation; native WebRTC and resource operations remain Experimental.

Vertex support

Install zhivex-ai-sdk[vertex]==0.28.3 for optional Google Auth ADC and refresh. Express API-key authentication is also supported. Partner routes, context caching, batch jobs, Interactions, Agent Runtime, Sessions, Memory Bank and RAG Engine are Beta native APIs. Model access, regional restrictions and incomplete service coverage remain explicit limits; old integration reports do not certify this wheel.

Native updates in 0.28.3

The direct catalog includes GPT-6 Sol/Luna, GPT-6.1 Sol, Claude Opus/Sonnet 5.5, Gemini 3.8 TTS and Qwen 3.7 rerank. New catalog entries do not inherit live certification.

Beta native APIs include OpenAI hosted multi-agent Responses, Gemini Voices, Qwen rerank and Qwen LiveTranslate. Review Claude 5.5 constraints. Ordinary OpenAI final-answer text is preserved when phase metadata is present; explicit hosted multi-agent requests filter the attributed root answer.