Compatibility matrix

What abliteration.ai supports across OpenAI Chat Completions, Responses, Anthropic Messages, and Count Tokens surfaces.

What abliteration.ai supports, grouped by API surface.

The image rows below apply to abliterated-model. abliterated-model-large and abliterated-model-large-v2 are text-only and reject image content with a 400 error. Video inputs are not supported on any model or API surface.

OpenAI — Chat Completions

POST /v1/chat/completions

Capability	Supported
Streaming	✓
Automatic prompt caching	✓
Cache routing hint (prompt_cache_key)	✓
Function calling	✓
Structured outputs (response_format: JSON mode, JSON Schema)	✓
Reasoning effort (reasoning_effort)	✓
Disable reasoning (reasoning_effort: "none")	✓; Large V2 maps this to hidden low reasoning
Hide reasoning trace (include_reasoning: false)	✓
Reasoning trace (message.reasoning_content)	✓
Web search	✓
Image inputs	✓
Video inputs	✗

OpenAI — Responses API

POST /v1/responses

Capability	Supported
Streaming	✓
Automatic prompt caching	✓
Cache routing hint (prompt_cache_key)	✓
Function calling	✓
Reasoning effort (reasoning.effort)	✓
Disable reasoning (reasoning.effort: "none")	✓; Large V2 maps this to hidden low reasoning
Reasoning trace (reasoning item in output[])	✓
Web search	✓
Image inputs	✓
Video inputs	✗

On Responses, the base model accepts effort up to xhigh (it rejects max); abliterated-model-large and abliterated-model-large-v2 accept up to max.

Anthropic — Messages API

POST /v1/messages

Capability	Supported
Streaming	✓
Automatic prompt caching	✓
Cache routing hint (x-abliteration-session-id)	✓
Function calling	✓
Reasoning effort (output_config.effort)	✓
Thinking budget (thinking.budget_tokens)	✓
Disable reasoning (thinking: false)	✓; Large V2 maps this to hidden low reasoning
Hide reasoning trace (include_reasoning: false, thinking.display: "omitted")	✓
Reasoning trace (thinking content block)	✓
Web search	✓
Web fetch	✓
Image inputs	✓
Video inputs	✗

Anthropic — Count Tokens

POST /v1/messages/count_tokens

Count input tokens before sending a request. See count tokens.

Policies

Published project policies apply on the standard /v1/chat/completions, /v1/messages, and /v1/responses endpoints. Add explicitly selected policies with policy_ids. See Policies in API requests.