Provider capability and compatibility matrix¶
This file is generated from providers.json. Do not edit it directly.
Run make docs after changing provider support.
Status meanings: supported is contract-tested for the adapter; model-dependent is selected through declared capabilities; unsupported is rejected or omitted.
| Capability | OpenAI | Anthropic |
|---|---|---|
| Completion | supported | supported |
| Streaming | supported | supported |
| Tool Definitions | supported | supported |
| Strict JSON Schema | model-dependent | model-dependent |
| Temperature | model-dependent | model-dependent |
| Completion Usage | supported | supported |
| Streaming Usage | supported | supported |
| Normalized Errors | supported | supported |
| Cancellation | supported | supported |
| Live Evaluation | supported | supported |
Adapter details¶
OpenAI¶
- Adapter:
treelang.ai.transport.OpenAITransport - Install:
treelang - Contract tests:
tests/test_testing.py - Guide:
docs/provider-capabilities.md - Official documentation
-
Checked model profiles:
-
gpt-4o: strict JSON Schema yes; temperature yes unknown-model: strict JSON Schema no; temperature no
Anthropic¶
- Adapter:
treelang.ai.anthropic.AnthropicTransport - Install:
treelang[anthropic] - Contract tests:
tests/test_testing.py - Guide:
docs/anthropic.md - Official documentation
-
Checked model profiles:
-
claude-sonnet-4-6: strict JSON Schema yes; temperature yes claude-3-5-sonnet: strict JSON Schema no; temperature yes
The matrix is validated in normal CI without credentials or network access. Live provider quality is measured separately by the credentialed evaluation workflow.