Protocol Conversion Compatibility

Troubleshooting

Encountering errors? See the API Error Codes Reference for a complete list of error types and troubleshooting steps.

开桥 lets you call the same model catalog through different API protocols. This page explains how common request parameters, response fields, and streaming events are mapped when requests are converted across protocols, and which fields do not have an equivalent in the target protocol.

This page is based on the conversion behavior in 开桥/rosetta-ai. rosetta-ai covers all 12 pairwise converters between OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Google Gemini.

Protocol Names

Short name	Protocol	开桥 API reference
Chat Completions	OpenAI Chat Completions	Create a Chat Completion
Responses	OpenAI Responses	Create a Model Response
Messages	Anthropic Messages	Create a Message
Gemini	Google Gemini / Vertex AI Generate Content	Generate Content

Support Overview

Read this table from the source protocol perspective. For example, the Chat Completions column shows where a field can be converted when your request starts in the OpenAI Chat Completions format.

Field category	Chat Completions	Messages	Responses	Gemini
Model `model`	All	All	All	All
System / instructions	All	All	All	All
`temperature`	All	All	All	All
`top_p`	All	All	All	All
`top_k`	Not supported	Gemini	Not supported	Messages
Max tokens	All	All	All	All
Stop sequences	Messages / Gemini	Chat Completions / Gemini	Not supported	Chat Completions / Messages
Function tools	All	All	All	All
`tool_choice`	All	All	All	All
Web search	All	All	All	All
Reasoning / thinking	All	All	All	All
JSON response format	All	All	All	All
Image input	Messages / Gemini	Chat Completions / Gemini	Gemini	Chat Completions / Messages
Documents / files	Messages / Gemini	Chat Completions / Gemini	Not supported	Chat Completions / Messages
Audio input	Gemini	Not supported	Not supported	Not supported
`seed`	Gemini	Not supported	Not supported	Chat Completions
`frequency_penalty`	Gemini	Not supported	Not supported	Chat Completions
`presence_penalty`	Gemini	Not supported	Not supported	Chat Completions
Candidate count `n` / `candidateCount`	Gemini	Not supported	Not supported	Chat Completions
Logprobs	Responses / Gemini	Not supported	Gemini	Chat Completions / Responses
Parallel tool calls	Messages / Responses	Chat Completions / Responses	Chat Completions / Messages	Not supported
Metadata / user	Messages / Responses	Chat Completions / Responses	Messages / Chat Completions	Not supported
`service_tier`	Messages / Responses	Chat Completions / Responses	Chat Completions / Messages	Not supported
Prompt cache key	Responses	Not supported	Chat Completions	Not supported
Streaming	All	All	All	All

Model support still applies

Protocol conversion only maps field shapes. It does not guarantee that every model supports the resulting capability. For exact support, check the supported parameters on the model detail page.

Chat Completions Request Conversion

Chat Completions → Messages

Chat Completions field	Messages field	Notes
`model`	`model`	Passed through
`messages` `system` / `developer`	`system`	Joined as text blocks
`messages` `user`	`messages[]` user	Text and images are supported; `file` content parts become an unsupported placeholder text
`messages` `assistant`	`messages[]` assistant	Text, `tool_use`, and thinking blocks
`messages` `tool`	`messages[]` user `tool_result`	Converted to tool result blocks
`max_tokens` / `max_completion_tokens`	`max_tokens`	Defaults to `4096` when omitted
`temperature`	`temperature`	Passed through
`top_p`	`top_p`	Passed through
`stop`	`stop_sequences`	String values become arrays
`tools` function	`tools` `input_schema`	JSON Schema becomes Anthropic tool schema
`web_search_options`	`tools` `web_search_20250305`	Converted to Anthropic web search tool
`tool_choice`	`tool_choice`	`auto` / `required` / `none` → `auto` / `any` / `none`; named functions become `tool` choices
`parallel_tool_calls`	`tool_choice.disable_parallel_tool_use`	Inverted semantics: `false` → `true`
`response_format` `json_schema`	`output_config.format`	JSON Schema output constraint
`response_format` `json_object`	`output_config.format`	JSON output format
`reasoning_effort`	`thinking`	`none` / `minimal` → disabled; `low` / `medium` / `high` → 2048 / 5120 / 10240 budget tokens
`user`	`metadata.user_id`	User identifier
`service_tier`	`service_tier`	Only `auto` / `standard_only` are forwarded
`stream`	`stream`	Passed through

Not converted: frequency_penalty, presence_penalty, seed, n, logprobs, top_logprobs, logit_bias.

Chat Completions → Responses

Chat Completions field	Responses field	Notes
`model`	`model`	Passed through
`messages`	`input`	Converted to Responses input items
`max_completion_tokens`	`max_output_tokens`	Max output tokens
`temperature`	`temperature`	Passed through
`top_p`	`top_p`	Passed through
`tools` function	`tools` function	Function tool definitions
`web_search_options`	`tools` `web_search`	Converted to Responses web search tool
`tool_choice`	`tool_choice`	Supports `auto` / `required` / `none` and named functions
`parallel_tool_calls`	`parallel_tool_calls`	Passed through
`response_format`	`text.format`	JSON / JSON Schema output format
`reasoning_effort`	`reasoning.effort`	Effort is passed through
`top_logprobs`	`include`	Requests logprobs-related output
`metadata`	`metadata`	Passed through
`service_tier`	`service_tier`	Passed through
`prompt_cache_key`	`prompt_cache_key`	Passed through
`prompt_cache_retention`	`prompt_cache_retention`	Passed through
`stream`	`stream` + `stream_options`	Streaming options

Not converted: frequency_penalty, presence_penalty, seed, n, logprobs, stop.

Chat Completions → Gemini

Chat Completions field	Gemini field	Notes
`model`	`model`	Passed through
`messages` `system` / `developer`	`config.systemInstruction`	System instruction
`messages` `user` / `assistant` / `tool`	`contents`	`assistant` becomes Gemini `model` role
`max_tokens` / `max_completion_tokens`	`config.maxOutputTokens`	Max output tokens
`temperature`	`config.temperature`	Passed through
`top_p`	`config.topP`	Field rename
`stop`	`config.stopSequences`	Stop sequences
`seed`	`config.seed`	Passed through
`frequency_penalty`	`config.frequencyPenalty`	Field rename
`presence_penalty`	`config.presencePenalty`	Field rename
`n`	`config.candidateCount`	Candidate count
`logprobs`	`config.responseLogprobs`	Whether to return logprobs
`top_logprobs`	`config.logprobs`	Number of top logprobs
`tools` function	`config.tools[].functionDeclarations`	Uses `parametersJsonSchema`
`web_search_options`	`config.tools[].googleSearch`	Converted to Google Search
`tool_choice`	`config.toolConfig.functionCallingConfig`	`auto` / `required` / `none` → `AUTO` / `ANY` / `NONE`; named functions use `allowedFunctionNames`
`response_format` `json_schema`	`config.responseMimeType` + `config.responseJsonSchema`	`responseMimeType` is `application/json`
`response_format` `json_object`	`config.responseMimeType`	`application/json`
`reasoning_effort`	`config.thinkingConfig`	`none` / `minimal` → `thinkingBudget: 0`; `low` / `medium` / `high` / `xhigh` → 2048 / 5120 / 10240 / 20480
`image_url` base64	`inlineData`	Image bytes
`image_url` URL	`fileData`	Image URL
`file` content part	`inlineData` / `fileData`	File input
`input_audio`	`inlineData`	Audio input

Not converted: logit_bias, user.

Messages Request Conversion

Messages → Chat Completions

Messages field	Chat Completions field	Notes
`model`	`model`	Passed through
`system`	`messages[]` system	String or `TextBlockParam[]` is joined as text
`messages` user	`messages[]` user	Text, image, document, and `tool_result`
`messages` assistant	`messages[]` assistant	Text, thinking, and `tool_use`
`max_tokens`	`max_completion_tokens`	Uses the newer field for reasoning models
`temperature`	`temperature`	Passed through
`top_p`	`top_p`	Passed through
`stop_sequences`	`stop`	Field rename
`tools` `input_schema`	`tools` function	JSON Schema tools
`tools` `web_search_20250305` / `web_search_20260209`	`web_search_options`	`max_uses` and `user_location` are mapped where possible
`tool_choice`	`tool_choice` + `parallel_tool_calls`	`auto` / `any` / `none` → `auto` / `required` / `none`; `disable_parallel_tool_use: true` → `parallel_tool_calls: false`
`output_config.format`	`response_format`	JSON / JSON Schema output format
`thinking`	`reasoning_effort`	disabled → `none`; enabled budget ≤2048 / ≤5120 / ≤10240 → `low` / `medium` / `high`
`metadata.user_id`	`user`	User identifier
`service_tier`	`service_tier`	`standard_only` → `default`
`stream`	`stream` + `stream_options`	Includes usage by default

Document block conversion:

source.type: "base64" / "url" → Chat Completions file content part
source.type: "text" → text content part
source.type: "content" → nested text / image blocks are unpacked

tool_result content conversion:

text is preserved
search_result inner text is flattened
tool_reference keeps the tool name
Images and documents are skipped because the OpenAI tool role only accepts text content

Not converted: top_k, Anthropic-specific metadata fields.

Messages → Responses

Messages field	Responses field	Notes
`model`	`model`	Passed through
`system`	`instructions`	Joined text
`messages`	`input`	Converted to Responses input items
`max_tokens`	`max_output_tokens`	Max output tokens
`temperature`	`temperature`	Passed through
`top_p`	`top_p`	Passed through
`tools` `input_schema`	`tools` function	Tool definitions
`tools` web search	`tools` web search	Converted to Responses web search tool
`tool_choice`	`tool_choice` + `parallel_tool_calls`	`auto` / `any` / `none` → `auto` / `required` / `none`
`thinking`	`reasoning`	Budget is mapped to effort
`output_config.format`	`text.format`	JSON Schema output format
`metadata.user_id`	`metadata.user_id`	Passed through
`service_tier`	`service_tier`	Passed through
`stream`	`stream`	Passed through

Not converted: stop_sequences, top_k.

Messages → Gemini

Messages field	Gemini field	Notes
`model`	`model`	Passed through
`system`	`config.systemInstruction`	String or text blocks
`messages` user	`contents[]` user	Text, image, document, and `tool_result`
`messages` assistant	`contents[]` model	Text, `tool_use`, and Gemini thought signature
`max_tokens`	`config.maxOutputTokens`	Max output tokens
`temperature`	`config.temperature`	Passed through
`top_p`	`config.topP`	Field rename
`top_k`	`config.topK`	Field rename
`stop_sequences`	`config.stopSequences`	Stop sequences
`tools` `input_schema`	`config.tools[].functionDeclarations`	Uses `parametersJsonSchema`
`tools` web search	`config.tools[].googleSearch`	Converted to Google Search
`tool_choice`	`config.toolConfig.functionCallingConfig`	`auto` / `any` / `none` → `AUTO` / `ANY` / `NONE`; named tools use `allowedFunctionNames`
`thinking`	`config.thinkingConfig`	disabled → `thinkingBudget: 0`; enabled → budget is passed through
`output_config.format`	`config.responseMimeType` + `config.responseJsonSchema`	JSON Schema output format

Not converted: metadata.

Document block conversion:

source.type: "base64" → inlineData
source.type: "url" → fileData
source.type: "text" → text
Other sources → unsupported placeholder text

Responses Request Conversion

Responses → Chat Completions

Responses field	Chat Completions field	Notes
`model`	`model`	Passed through
`instructions`	`messages[]` system	System instruction
`input`	`messages`	Converted to Chat Completions messages
`max_output_tokens`	`max_completion_tokens`	Max output tokens
`temperature`	`temperature`	Passed through
`top_p`	`top_p`	Passed through
`tools` function	`tools` function	Tool definitions
`tools` web search	`web_search_options`	Converted to Chat Completions web search options
`tool_choice`	`tool_choice`	Supports `auto` / `required` / `none` and named functions
`parallel_tool_calls`	`parallel_tool_calls`	Passed through
`reasoning.effort`	`reasoning_effort`	Passed through
`text.format`	`response_format`	JSON / JSON Schema output format
`text.verbosity`	`verbosity`	Passed as an extension field
`metadata`	`metadata`	Passed through
`service_tier`	`service_tier`	Passed through
`prompt_cache_key`	`prompt_cache_key`	Passed through
`prompt_cache_retention`	`prompt_cache_retention`	Passed through
`safety_identifier`	`safety_identifier`	Passed through
`include` logprobs	`top_logprobs`	Set to `20`
`stream`	`stream` + `stream_options`	Streaming options

Not converted: previous_response_id.

Responses → Messages

Responses field	Messages field	Notes
`model`	`model`	Passed through
`instructions`	`system`	System instruction
`input`	`messages`	Converted to Anthropic messages
`max_output_tokens`	`max_tokens`	Defaults to `4096` when omitted
`temperature`	`temperature`	Passed through
`top_p`	`top_p`	Passed through
`tools` function	`tools` `input_schema`	Tool definitions
`tools` web search	`tools` `web_search_20250305`	Converted to Anthropic web search tool
`tool_choice` + `parallel_tool_calls`	`tool_choice`	`auto` / `required` / `none` → `auto` / `any` / `none`; `parallel_tool_calls: false` → `disable_parallel_tool_use: true`
`reasoning.effort`	`thinking`	`low` / `medium` / `high` → 2048 / 5120 / 10240 budget tokens
`text.format` `json_schema`	`output_config.format`	JSON Schema output format
`metadata.user_id`	`metadata.user_id`	Passed through
`service_tier`	`service_tier`	Only `auto` / `standard_only` are forwarded
`stream`	`stream`	Passed through

Not converted: previous_response_id, prompt_cache_key, prompt_cache_retention, safety_identifier.

Responses → Gemini

Responses field	Gemini field	Notes
`model`	`model`	Passed through
`instructions`	`config.systemInstruction`	System instruction
`input`	`contents`	Converted to Gemini contents
`max_output_tokens`	`config.maxOutputTokens`	Max output tokens
`temperature`	`config.temperature`	Passed through
`top_p`	`config.topP`	Field rename
`tools` function	`config.tools[].functionDeclarations`	Uses `parametersJsonSchema`
`tools` web search	`config.tools[].googleSearch`	Converted to Google Search
`tool_choice`	`config.toolConfig.functionCallingConfig`	`auto` / `required` / `none` → `AUTO` / `ANY` / `NONE`; named functions use `allowedFunctionNames`
`reasoning.effort`	`config.thinkingConfig`	`low` / `medium` / `high` → 2048 / 5120 / 10240 budget
`text.format`	`config.responseMimeType` + `config.responseJsonSchema`	JSON / JSON Schema output format
`include` logprobs	`config.responseLogprobs` + `config.logprobs`	Logprobs output

Not converted: metadata, previous_response_id, parallel_tool_calls, prompt_cache_key, prompt_cache_retention, safety_identifier.

Gemini Request Conversion

Gemini → Chat Completions

Gemini field	Chat Completions field	Notes
`model`	`model`	Passed through
`config.systemInstruction`	`messages[]` system	System instruction
`contents`	`messages`	`user` / `model` → `user` / `assistant`
`config.maxOutputTokens`	`max_completion_tokens`	Max output tokens
`config.temperature`	`temperature`	Passed through
`config.topP`	`top_p`	Field rename
`config.stopSequences`	`stop`	Stop sequences
`config.seed`	`seed`	Passed through
`config.frequencyPenalty`	`frequency_penalty`	Field rename
`config.presencePenalty`	`presence_penalty`	Field rename
`config.candidateCount`	`n`	Candidate count
`config.responseLogprobs`	`logprobs`	Whether to return logprobs
`config.logprobs`	`top_logprobs`	Number of top logprobs
`config.tools[].functionDeclarations`	`tools` function	Tool definitions
`config.tools[].googleSearch`	`web_search_options`	Web search options
`config.toolConfig`	`tool_choice`	`AUTO` / `ANY` / `NONE` → `auto` / `required` / `none`; single `allowedFunctionNames` becomes a named function
`config.responseMimeType` + `responseJsonSchema`	`response_format`	JSON / JSON Schema output format
`config.thinkingConfig`	`reasoning_effort`	`thinkingBudget: 0` → `none`; ≤2048 / ≤5120 / >5120 → `low` / `medium` / `high`

Not converted: config.safetySettings, config.cachedContent, config.responseModalities.

Gemini → Messages

Gemini field	Messages field	Notes
`model`	`model`	Passed through
`config.systemInstruction`	`system`	System instruction
`contents`	`messages`	`user` / `model` → `user` / `assistant`
`config.maxOutputTokens`	`max_tokens`	Defaults to `4096` when omitted
`config.temperature`	`temperature`	Passed through
`config.topP`	`top_p`	Field rename
`config.topK`	`top_k`	Field rename
`config.stopSequences`	`stop_sequences`	Stop sequences
`config.tools[].functionDeclarations`	`tools` `input_schema`	Tool definitions
`config.tools[].googleSearch`	`tools` `web_search_20250305`	Converted to Anthropic web search tool
`config.toolConfig`	`tool_choice`	`AUTO` / `ANY` / `NONE` → `auto` / `any` / `none`; single `allowedFunctionNames` becomes a named tool
`config.thinkingConfig`	`thinking`	Budget is passed through; `0` means disabled
`config.responseMimeType` + `responseJsonSchema`	`output_config`	JSON Schema output format

Not converted: config.seed, config.frequencyPenalty, config.presencePenalty, config.candidateCount, config.safetySettings, config.cachedContent.

Gemini → Responses

Gemini field	Responses field	Notes
`model`	`model`	Passed through
`config.systemInstruction`	`instructions`	System instruction
`contents`	`input`	Converted to Responses input items
`config.maxOutputTokens`	`max_output_tokens`	Max output tokens
`config.temperature`	`temperature`	Passed through
`config.topP`	`top_p`	Field rename
`config.tools[].functionDeclarations`	`tools` function	Tool definitions
`config.tools[].googleSearch`	`tools` `web_search_preview`	Converted to Responses web search tool
`config.toolConfig`	`tool_choice`	`AUTO` / `ANY` / `NONE` → `auto` / `required` / `none`
`config.thinkingConfig`	`reasoning`	Budget is mapped to effort
`config.responseMimeType` + `responseJsonSchema`	`text.format`	JSON / JSON Schema output format

Not converted: config.seed, config.frequencyPenalty, config.presencePenalty, config.candidateCount, config.safetySettings, config.cachedContent.

Response Field Conversion

Source response	Target response	Content field	Tool call field	Reasoning field	Finish reason
Messages	Chat Completions	`content[]` text → `choices[0].message.content`	`tool_use` → `tool_calls`	`thinking` → `reasoning` + `reasoning_details`	`end_turn` / `tool_use` / `max_tokens` / `refusal` → `stop` / `tool_calls` / `length` / `content_filter`
Responses	Chat Completions	`output[]` message output_text → `choices[0].message.content`	`function_call` → `tool_calls`	`reasoning` → `reasoning` + `reasoning_details`	`completed` / `incomplete` / `failed` → `stop` / `length` / `stop`; content filter becomes `content_filter`
Gemini	Chat Completions	`parts[].text` → `choices[0].message.content`	`functionCall` → `tool_calls`	`thought` → `reasoning` + `reasoning_details`	`STOP` / `MAX_TOKENS` / `SAFETY` → `stop` / `length` / `content_filter`
Chat Completions	Messages	`message.content` → `content[]` text	`tool_calls` → `tool_use`	`reasoning` → `thinking`	`stop` / `tool_calls` / `length` / `content_filter` → `end_turn` / `tool_use` / `max_tokens` / `refusal`
Responses	Messages	`output[]` message → `content[]` text	`function_call` → `tool_use`	`reasoning` → `thinking`	`completed` / `incomplete` / `failed` → `end_turn` / `max_tokens` / `end_turn`
Gemini	Messages	`parts[].text` → `content[]` text	`functionCall` → `tool_use`	`thought` → `thinking`	`STOP` / `MAX_TOKENS` / `SAFETY` → `end_turn` / `max_tokens` / `refusal`
Chat Completions	Responses	`message.content` → `output[]` message output_text	`tool_calls` → `function_call`	`reasoning` → `output[]` reasoning	`stop` / `tool_calls` / `length` / `content_filter` → `completed` / `completed` / `incomplete` / `failed`
Messages	Responses	`content[]` text → `output[]` message output_text	`tool_use` → `function_call`	`thinking` → `output[]` reasoning	`end_turn` / `max_tokens` / `refusal` → `completed` / `incomplete` / `failed`
Gemini	Responses	`parts[].text` → `output[]` message output_text	`functionCall` → `function_call`	`thought` → `output[]` reasoning	`STOP` / `MAX_TOKENS` / `SAFETY` → `completed` / `incomplete` / `failed`
Chat Completions	Gemini	`message.content` → `parts[].text`	`tool_calls` → `functionCall`	`reasoning` → `parts[].thought`	`stop` / `tool_calls` / `length` / `content_filter` → `STOP` / `STOP` / `MAX_TOKENS` / `SAFETY`
Messages	Gemini	`content[]` text → `parts[].text`	`tool_use` → `functionCall`	`thinking` → `parts[].thought`	`end_turn` / `max_tokens` / `refusal` → `STOP` / `MAX_TOKENS` / `SAFETY`
Responses	Gemini	`output[]` message output_text → `parts[].text`	`function_call` → `functionCall`	`reasoning` → `parts[].thought`	`completed` / `incomplete` / `failed` → `STOP` / `MAX_TOKENS` / `SAFETY`

Usage Field Conversion

Source protocol	Target protocol	Main mapping
Messages	Chat Completions	`usage.input_tokens` → `prompt_tokens`; `usage.output_tokens` → `completion_tokens`; `cache_read_input_tokens` → `prompt_tokens_details.cached_tokens`; `server_tool_use.web_search_requests` → `prompt_tokens_details.web_search`
Responses	Chat Completions	`usage.input_tokens` / `output_tokens` / `total_tokens` are mapped directly; cached tokens and reasoning tokens are mapped to OpenAI-style detail fields
Gemini	Chat Completions	`promptTokenCount + toolUsePromptTokenCount` → `prompt_tokens`; `candidatesTokenCount + thoughtsTokenCount` → `completion_tokens`; `cachedContentTokenCount` → cached tokens; `thoughtsTokenCount` → reasoning tokens
Chat Completions	Messages	`prompt_tokens` → `input_tokens`; `completion_tokens` → `output_tokens`; cached tokens and web search counts map to Anthropic usage extension fields
Responses	Messages	cached tokens → `cache_read_input_tokens`; web search calls are counted in `server_tool_use`
Gemini	Messages	`promptTokenCount + toolUsePromptTokenCount` → `input_tokens`; `candidatesTokenCount + thoughtsTokenCount` → `output_tokens`; `cachedContentTokenCount` → `cache_read_input_tokens`
Chat Completions	Responses	cached tokens and reasoning tokens are mapped to Responses usage; web search calls are counted
Messages	Responses	`input_tokens` / `output_tokens` are mapped directly; `cache_read_input_tokens` → cached tokens
Gemini	Responses	Gemini `usageMetadata` is mapped to Responses usage; tool-use prompt tokens and thoughts tokens are folded into the relevant counts
Chat Completions	Gemini	cached tokens → `cachedContentTokenCount`; reasoning tokens → `thoughtsTokenCount`
Messages	Gemini	`cache_read_input_tokens` → `cachedContentTokenCount`
Responses	Gemini	cached tokens → `cachedContentTokenCount`; reasoning tokens → `thoughtsTokenCount`

Streaming Conversion

All 12 converters support streaming. Event names differ by protocol, but these content types are preserved where possible:

Content type	Conversion behavior
Text deltas	Preserved as the target protocol's text delta
Tool call deltas	Tool name, tool call ID, and argument deltas are preserved; when the target protocol has no exact event shape, the closest target event is emitted
Reasoning / thinking deltas	`reasoning`, `thinking_delta`, and Gemini thought events are converted
Web search / grounding	Responses web search, Anthropic `web_search_tool_result`, Gemini grounding metadata, and Chat Completions annotations are mapped where possible
Usage	If upstream streaming includes a trailing usage payload, it is converted into the target protocol's usage event or final chunk fields

Common Differences And Notes

Responses has no direct request field equivalent for stop / stop_sequences; those fields are not converted when going from Chat Completions or Messages to Responses.
Gemini does not support parallel_tool_calls, metadata, service_tier, prompt cache keys, or several OpenAI / Anthropic extension parameters.
Anthropic Messages top_k is preserved only when converting to Gemini. It is not converted to Chat Completions or Responses.
Chat Completions seed, frequency_penalty, presence_penalty, and n mainly map to Gemini. They have no equivalent in Messages / Responses.
reasoning_effort and thinking are range mappings, not exact token-limit preservation. If you need precise thinking budget control, prefer the target protocol's native parameter.
Images and documents inside Anthropic tool_result blocks are skipped when converting to the Chat Completions tool role because the OpenAI tool role accepts text content only.
Protocol conversion preserves web search / grounding data where possible, but citation field names and nesting differ across protocols.

Protocol Conversion Compatibility ​

Protocol Names ​

Support Overview ​

Chat Completions Request Conversion ​

Chat Completions → Messages ​

Chat Completions → Responses ​

Chat Completions → Gemini ​

Messages Request Conversion ​

Messages → Chat Completions ​

Messages → Responses ​

Messages → Gemini ​

Responses Request Conversion ​

Responses → Chat Completions ​

Responses → Messages ​

Responses → Gemini ​

Gemini Request Conversion ​

Gemini → Chat Completions ​

Gemini → Messages ​

Gemini → Responses ​

Response Field Conversion ​

Usage Field Conversion ​

Streaming Conversion ​

Common Differences And Notes ​