This page compares request parameters across the Runars endpoints. If a parameter appears in multiple endpoints, it may have different names or shapes depending on the API dialect.

For authoritative details on any parameter, see the dedicated page for the endpoint you're using:

/v1/messages — Anthropic dialect
/v1/chat/completions — OpenAI dialect
/v1/responses — OpenAI Responses (simplified)
/v1/messages/count_tokens — Token counting

Parameter comparison table

Concept	`/v1/messages`	`/v1/chat/completions`	`/v1/responses`	`/v1/count_tokens`
Model	`model`	`model`	`model`	`model`
Max tokens	`max_tokens` (required)	`max_tokens` (optional)	`max_tokens` (optional)	—
Input/Prompt	`messages` array	`messages` array	`input` string	`messages` array
System/Instructions	`system` string	System role message in `messages`	`instructions` string	`system` string
Streaming	`stream` boolean	`stream` boolean	`stream` boolean	—
Temperature	—	`temperature` (0–2)	—	—
Top P	—	`top_p` (0–1)	—	—
Tool calling	`tools` array `tool_choice`	`tools` array `tool_choice`	❌ Not supported	—
Stream options	—	`stream_options` object	—	—

Key differences

Input format

Anthropic (/v1/messages): Uses a messages array of objects with role and content.
OpenAI (/v1/chat/completions): Uses a messages array of objects with role and content.
Responses (/v1/responses): Simplified to a single input string (no history support).

System prompt / Instructions

Anthropic: Dedicated system parameter (string).
OpenAI Chat Completions: No dedicated parameter; use a message with role: "system" in the messages array.
Responses: Dedicated instructions parameter (string).

Sampling

Anthropic: No temperature/top_p (model uses default sampling).
OpenAI Chat Completions: Supports temperature (0–2, default 1) and top_p (0–1).
Responses: No sampling parameters.

Tool calling

Anthropic (/v1/messages): Supports tools array and tool_choice parameter.
OpenAI Chat Completions (/v1/chat/completions): Supports tools array and tool_choice parameter.
Responses (/v1/responses): ⚠️ Not yet supported. Use one of the other endpoints if you need tool calling.

Streaming

All three endpoints support streaming via a stream: true parameter, but response chunk formats differ:

Anthropic: Server-sent events with message_start, message_delta, message_stop events.
OpenAI Chat Completions: Server-sent events with chat.completion.chunk delta format.
Responses: Server-sent events with custom response.* event types.

Max tokens requirement

Anthropic (/v1/messages): max_tokens is required (1–8000).
OpenAI Chat Completions (/v1/chat/completions): max_tokens is optional (default: unlimited).
Responses (/v1/responses): max_tokens is optional.
Count Tokens (/v1/count_tokens): Does not accept max_tokens (it only counts, not generates).

Parameters Reference