DOCS_IMPROVEMENT_CHECKLIST.md

Crazyrouter Docs Improvement Checklist

Last updated: 2026-03-23

Rules

• Update this file immediately after each completed item.
• Keep one source of truth for scope, status, and completion notes.
• Mark completed items with [x], in-progress items with [-], pending items with [ ].

Current Status

• Initialize this checklist and completion log
• Fix factual inconsistencies in existing Mintlify docs
• Add missing docs for already-supported backend endpoints
• Build a standard guide template for application/integration docs
• Expand OpenClaw into a full usage guide
• Upgrade priority application guides
• Upgrade secondary application guides
• Add production-verified unified capability pages

A. Fix Existing Doc Inconsistencies

• Remove or correct POST /api/upload in upload.mdx and api-endpoint.mdx
• Replace GET /api/token/models with the real supported model-list flow
• Correct /api/user/self auth requirements in token/account docs
• Correct Realtime auth docs to remove unsupported api_key query usage
• Align token CRUD docs with real request/response fields
• Verify and correct Sora characters documentation
• Verify and correct Replicate native Flux documentation

B. Add Docs For Supported But Undocumented Endpoints

• Keep GET /api/status/page out of public docs; it is a page-data endpoint behind the user-facing /status page
• Keep GET /api/status/health out of public docs; it is an ops-facing health endpoint
• Keep GET /healthz/live and GET /healthz/ready out of public docs; they are deployment probes
• Add Claude token counting doc for POST /v1/messages/count_tokens
• Add Gemini compatible model-list doc for GET /v1beta/openai/models
• Add video alias doc for POST /v1/video/generations
• Add video alias doc for GET /v1/video/generations/{task_id}
• Add Ali Bailian video doc

C. Standardize Application Guide Structure

• Define a common structure for all application guides
• Define a common troubleshooting section format
• Define a common verification checklist for first successful setup
• Define recommended token/model guidance format

D. OpenClaw Guide Expansion

• Rewrite page structure from install note to full usage guide
• Add post-install first-run verification steps
• Add config file locations and key fields
• Add Crazyrouter token best practices for OpenClaw
• Add model recommendation table by use case
• Add IM integration walkthroughs
• Add service management, restart, and logs section
• Add upgrade guide
• Add uninstall guide
• Add troubleshooting matrix

E. Priority Application Guides

• Upgrade integrations/cursor.mdx
• Upgrade integrations/claude-code.mdx
• Upgrade integrations/codex.mdx
• Upgrade integrations/cline.mdx
• Upgrade integrations/aider.mdx

F. Secondary Application Guides

• Upgrade integrations/cherry-studio.mdx
• Upgrade integrations/lobechat.mdx
• Upgrade integrations/nextchat.mdx
• Upgrade integrations/dify.mdx
• Upgrade integrations/n8n.mdx
• Review remaining integration pages and batch-upgrade them

G. Verification

• Verify key route and auth assumptions against the local Docker test environment on port 4000
• Tighten n8n setup guidance to reflect version-dependent Base URL placement and keep zh/en docs in sync
• Revalidate Cherry Studio guide against official provider docs and keep root-domain guidance as-is
• Revalidate LobeChat guide against official provider docs and keep /v1 proxy guidance as-is
• Revalidate NextChat setup wording against official docs and the local 4000 status payload
• Revalidate Dify provider guidance against official docs and clarify admin-role requirements
• Add a second-pass Cherry Studio note set for Manage / + model enablement and full-path # edge cases
• Add a second-pass LobeChat note set for /v1 proxy duplication and self-hosted model pinning
• Add a second-pass NextChat note set for env-only deployments and minimal first-pass validation
• Add a second-pass Dify note set for provider-label variance and one-model-first validation
• Sync verify_docs_api.py with the revised public-doc scope for token-management pages

H. Public-Surface Audit

• Add a public-facing https://crazyrouter.com/status link in the introduction pages
• Remove GET /api/user/self account-info docs from public docs; keep it as a dashboard/internal user endpoint
• Relabel token-management navigation as API-key / console-facing docs instead of core model-calling APIs
• Add explicit dashboard-auth and non-sk-xxx positioning notes to the remaining /api/token/* public docs

I. Deepen Coding Tool Guides

• Expand integrations/claude-code.mdx and en/integrations/claude-code.mdx with Git / Node install commands, persistent env vars, repo snapshot, and first-run validation
• Further split integrations/claude-code.mdx and en/integrations/claude-code.mdx into explicit Windows vs macOS install paths, verification commands, and env-var checks
• Expand integrations/codex.mdx and en/integrations/codex.mdx with Git / Node install commands, config.toml writing steps, persistent env vars, repo snapshot, and first-run validation
• Further split integrations/codex.mdx and en/integrations/codex.mdx into explicit Windows vs macOS install paths, verification commands, and config-file checks
• Expand integrations/aider.mdx and en/integrations/aider.mdx with Git / Python install commands, official installer steps, persistent env vars, repo snapshot, and first-run validation
• Further refine integrations/aider.mdx and en/integrations/aider.mdx with explicit Windows vs macOS install paths, command-path verification, and Crazyrouter raw model ID usage
• Expand integrations/cline.mdx and en/integrations/cline.mdx with Git / VS Code / extension install steps, repo snapshot guidance, and first-run validation
• Further refine integrations/cline.mdx and en/integrations/cline.mdx with explicit Windows vs macOS install paths, code command verification, UI-based provider setup notes, and VS Code proxy guidance
• Expand integrations/langchain.mdx and en/integrations/langchain.mdx with Python / JS install steps, env-var persistence, minimal validation, embeddings, and RAG rollout guidance
• Expand integrations/immersive-translate.mdx and en/integrations/immersive-translate.mdx with full custom-URL setup, model guidance, rate-limit notes, and first-pass translation validation
• Expand integrations/coze.mdx and en/integrations/coze.mdx with public-safe HTTP plugin / workflow guidance, enterprise custom-model boundary notes, staged validation, and troubleshooting
• Expand integrations/chatbox.mdx and en/integrations/chatbox.mdx with official-provider-aligned host/path guidance, model rollout order, knowledge-base notes, and troubleshooting
• Expand integrations/zotero.mdx and en/integrations/zotero.mdx with plugin-selection guidance, custom-endpoint boundary notes, staged academic validation, and troubleshooting

J. Information Architecture

• Add top-level 核心能力 / Features navigation groups to surface tool calling, structured outputs, web search, reasoning, vision, and PDF-related pages without forcing users to choose a protocol first

K. Unified Capability Pages

• Add features/tool-calling.mdx and en/features/tool-calling.mdx
• Add features/structured-outputs.mdx and en/features/structured-outputs.mdx
• Add features/web-search.mdx and en/features/web-search.mdx
• Add features/reasoning.mdx and en/features/reasoning.mdx
• Revalidate the four capability pages against Crazyrouter production on 2026-03-22 and narrow claims to reproducible response markers

L. Reasoning Protocol Cleanup

• Rework chat/openai/reasoning.mdx and en/chat/openai/reasoning.mdx to separate Chat reasoning_effort from Responses-visible reasoning
• Rework chat/responses/gpt5-thinking.mdx and en/chat/responses/gpt5-thinking.mdx to use production-verified gpt-5.4 response shapes and stream events
• Rework reference/thinking.mdx and en/reference/thinking.mdx to document protocol-specific thinking fields instead of outdated cross-protocol assumptions

M. Protocol Page Refresh

• Rework chat/openai/web-search.mdx and en/chat/openai/web-search.mdx to document the currently verified Responses-based web-search path
• Rework chat/anthropic/messages.mdx and en/chat/anthropic/messages.mdx around claude-sonnet-4-6 and claude-opus-4-6-thinking
• Rework chat/gemini/native.mdx and en/chat/gemini/native.mdx around gemini-3-pro-preview and currently verified native capabilities

N. Overview Page Refresh

• Rework chat/gemini/openai-compat.mdx and en/chat/gemini/openai-compat.mdx around gemini-3-pro-preview
• Rework chat/responses/overview.mdx and en/chat/responses/overview.mdx around the currently verified gpt-5.4 Responses behavior
• Rework chat/openai/overview.mdx and en/chat/openai/overview.mdx to describe the stable Chat Completion object shape without overclaiming optional fields

O. Model List And Tutorial Refresh

• Rework chat/openai/models.mdx and en/chat/openai/models.mdx around the 2026-03-23 live /v1/models response
• Rework chat/gemini/openai-models.mdx and en/chat/gemini/openai-models.mdx around the 2026-03-23 live /v1beta/openai/models response
• Rework reference/tutorial.mdx and en/reference/tutorial.mdx into production-verified starter paths instead of generic compatibility claims

P. Endpoint And Special-Model Refresh

• Rework chat/openai/special-models.mdx and en/chat/openai/special-models.mdx around the 2026-03-23 live availability check
• Rework chat/openai/completions.mdx and en/chat/openai/completions.mdx around the current gpt-5.4 chat object and stream behavior
• Rework chat/anthropic/overview.mdx and en/chat/anthropic/overview.mdx around claude-sonnet-4-6 and the current native SSE event sequence

Q. Remaining Legacy And Multimodal Cleanup

• Rework chat/gemini/tools.mdx and en/chat/gemini/tools.mdx around the 2026-03-23 live gemini-3-pro-preview tool checks
• Rework chat/gemini/document.mdx and en/chat/gemini/document.mdx around the 2026-03-23 live PDF + schema checks
• Rework chat/openai/image-creation.mdx and en/chat/openai/image-creation.mdx around the 2026-03-23 observed chat-image response shape and the verified gpt-image-1 fallback path
• Rework chat/openai/completions-legacy.mdx and en/chat/openai/completions-legacy.mdx around the 2026-03-23 legacy-model availability result
• Rework chat/gemini/image-gen.mdx and en/chat/gemini/image-gen.mdx around the 2026-03-23 observed per-model output-shape differences
• Rework chat/gemini/image-edit.mdx and en/chat/gemini/image-edit.mdx into current availability notes because same-day production retests were unavailable or timed out

R. Responses And Vision Refresh

• Rework chat/responses/create.mdx and en/chat/responses/create.mdx around the 2026-03-23 live gpt-5.4 basic and streaming checks
• Rework chat/responses/function-calling.mdx and en/chat/responses/function-calling.mdx around the 2026-03-23 live function_call output
• Rework chat/responses/web-search.mdx and en/chat/responses/web-search.mdx around the 2026-03-23 live web_search_call output
• Rework chat/openai/structured-output.mdx and en/chat/openai/structured-output.mdx around the 2026-03-23 live gpt-5.4 json_schema result
• Rework chat/openai/function-calling.mdx and en/chat/openai/function-calling.mdx around the 2026-03-23 live gpt-5.4 tool_calls result
• Rework chat/openai/vision.mdx and en/chat/openai/vision.mdx around the 2026-03-23 live gpt-5.4 data-URL image input
• Rework chat/anthropic/vision.mdx, chat/anthropic/pdf.mdx, and chat/anthropic/completions.mdx plus English counterparts around the 2026-03-23 live claude-sonnet-4-6 checks
• Rework chat/gemini/vision.mdx and en/chat/gemini/vision.mdx around the 2026-03-23 live gemini-3-pro-preview image-input check

S. Entry-Page Example Refresh

• Update quickstart.mdx, making-requests.mdx, and authentication.mdx plus English counterparts to use the 2026-03-23 production-verified gpt-5.4 starter example

Completion Log

• 2026-03-18: Created crazyrouter-docs/DOCS_IMPROVEMENT_CHECKLIST.md and initialized the tracking workflow.
• 2026-03-18: Corrected unsupported public upload documentation in upload.mdx, en/upload.mdx, api-endpoint.mdx, and en/api-endpoint.mdx.
• 2026-03-18: Rewrote token model-list docs to use the real public endpoint GET /v1/models in token-management/models.mdx and en/token-management/models.mdx.
• 2026-03-18: Corrected /api/user/self auth guidance in token-management/account.mdx and en/token-management/account.mdx.
• 2026-03-18: Corrected Realtime authentication guidance in audio/realtime.mdx and en/audio/realtime.mdx.
• 2026-03-18: Aligned token CRUD docs with current request/response fields in token-management/create|list|update|search.mdx and their English counterparts.
• 2026-03-18: Removed unsupported public-route claims from video/sora.mdx, en/video/sora.mdx, images/flux.mdx, and en/images/flux.mdx.
• 2026-03-18: Added crazyrouter-docs/APPLICATION_GUIDE_TEMPLATE.md to standardize structure, troubleshooting, verification, and token/model guidance for application docs.
• 2026-03-18: Rewrote openclaw-deploy.mdx and en/openclaw-deploy.mdx into full usage guides covering setup, verification, config, model selection, IM integration, operations, upgrades, uninstall, and troubleshooting.
• 2026-03-18: Upgraded integrations/cursor.mdx and en/integrations/cursor.mdx into full setup guides with BYOK limitations, token strategy, model recommendations, verification, and troubleshooting.
• 2026-03-18: Upgraded integrations/claude-code.mdx and en/integrations/claude-code.mdx into full setup guides covering Anthropic-native setup, token strategy, model recommendations, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/codex.mdx and en/integrations/codex.mdx into full setup guides covering custom provider setup, Responses API routing, token strategy, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/cline.mdx and en/integrations/cline.mdx into full setup guides covering OpenAI Compatible provider setup, agent permission strategy, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/aider.mdx and en/integrations/aider.mdx into full setup guides covering OpenAI-compatible setup, config file strategy, model recommendations, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/cherry-studio.mdx and en/integrations/cherry-studio.mdx into full desktop usage guides covering provider setup, model management, MCP rollout order, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/lobechat.mdx and en/integrations/lobechat.mdx into full setup guides covering OpenAI proxy setup, self-host deployment defaults, model strategy, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/nextchat.mdx and en/integrations/nextchat.mdx into full setup guides covering root-domain base URL usage, self-host environment variables, model rollout, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/dify.mdx and en/integrations/dify.mdx into full platform guides covering model-provider setup, chat-first rollout, embedding strategy, environment isolation, validation, and troubleshooting.
• 2026-03-18: Upgraded integrations/n8n.mdx and en/integrations/n8n.mdx into full automation guides covering OpenAI credentials, minimal workflows, AI Agent rollout, HTTP Request fallback, validation, and troubleshooting.
• 2026-03-18: Verified key documented routes and auth patterns against the local Docker test environment on http://127.0.0.1:4000, including /healthz/live, /api/status/health, /api/status/page, GET /v1/models, POST /v1/messages/count_tokens, GET /v1beta/openai/models, POST /v1/responses, and video task routes.
• 2026-03-18: Synced en/integrations/n8n.mdx with the updated Chinese guide and clarified that Base URL may be credential-level or node-level depending on n8n version, with node-level setup preferred for first-pass Crazyrouter validation.
• 2026-03-18: Rechecked integrations/cherry-studio.mdx and en/integrations/cherry-studio.mdx against Cherry Studio's current provider docs; kept the root-domain API URL guidance as-is because Cherry Studio documents root-address entry for OpenAI-compatible upstreams and model management via Manage or manual add.
• 2026-03-18: Rechecked integrations/lobechat.mdx and en/integrations/lobechat.mdx against current LobeChat provider docs; kept the /v1 proxy guidance as-is because LobeChat documents OPENAI_PROXY_URL as the OpenAI request base URL and shows /v1-style defaults/examples.
• 2026-03-18: Tightened integrations/nextchat.mdx and en/integrations/nextchat.mdx to justify root-domain BASE_URL guidance with the official NextChat docs pattern plus the local http://127.0.0.1:4000/api/status server_address, instead of relying on an implicit /v1 implementation claim.
• 2026-03-18: Tightened integrations/dify.mdx and en/integrations/dify.mdx to align with official Dify provider docs, clarifying that Crazyrouter should be entered as a custom OpenAI-compatible upstream via API Base URL and that Model Provider configuration typically requires workspace-admin or owner permissions.
• 2026-03-18: Added a second-pass practical note set to integrations/cherry-studio.mdx and en/integrations/cherry-studio.mdx, covering root-URL auto-path behavior, when the special trailing-# mode is actually needed, and the common Manage-list vs enabled-model-list confusion during first validation.
• 2026-03-18: Added a second-pass practical note set to integrations/lobechat.mdx and en/integrations/lobechat.mdx, covering /v1 proxy reasoning, duplicate-suffix failures behind custom reverse proxies, and a tighter self-hosted starter model list via OPENAI_MODEL_LIST.
• 2026-03-18: Added a second-pass practical note set to integrations/nextchat.mdx and en/integrations/nextchat.mdx, clarifying that some deployments expose settings only through env vars and that the minimal first-pass validation path should stay API Key + root BASE_URL + gpt-4.1.
• 2026-03-18: Added a second-pass practical note set to integrations/dify.mdx and en/integrations/dify.mdx, clarifying provider-label / plugin variance across Dify versions and reinforcing the one-provider, one-LLM, one-chat-app validation sequence before adding embeddings or rerank.
• 2026-03-18: Synced verify_docs_api.py with the revised public-doc scope by removing the withdrawn GET /api/user/self doc check, relabeling the token-management verification section as console/API-key scope, and fixing token-list count extraction to match the current paginated response shape.
• 2026-03-18: Added status-page docs in reference/status-page.mdx and en/reference/status-page.mdx, documenting the public GET /api/status/page aggregate payload for system state, traffic, response-time trends, and service health; later withdrawn from public docs after deciding /status is the correct user-facing surface.
• 2026-03-18: Added health-check docs in reference/health-check.mdx and en/reference/health-check.mdx, documenting the public GET /api/status/health endpoint and how it differs from readiness probes; later withdrawn from public docs because it is ops-facing.
• 2026-03-18: Added infrastructure probe docs in reference/infra-probes.mdx and en/reference/infra-probes.mdx, covering GET /healthz/live and GET /healthz/ready for liveness/readiness probing; later withdrawn from public docs because they are deployment probes.
• 2026-03-18: Added Claude token-counting docs in chat/anthropic/count-tokens.mdx and en/chat/anthropic/count-tokens.mdx, documenting POST /v1/messages/count_tokens, its auth pattern, estimate semantics, and image/file warning behavior.
• 2026-03-18: Added Gemini-compatible model-list docs in chat/gemini/openai-models.mdx and en/chat/gemini/openai-models.mdx, documenting GET /v1beta/openai/models and its Gemini-style auth with OpenAI-style list semantics.
• 2026-03-18: Added video alias creation docs in video/openai-video-alias-create.mdx and en/video/openai-video-alias-create.mdx, documenting POST /v1/video/generations as a compatibility route and noting the verified local business-error behavior when a model is unavailable.
• 2026-03-18: Added video alias query docs in video/openai-video-alias-query.mdx and en/video/openai-video-alias-query.mdx, documenting GET /v1/video/generations/{task_id} and the verified task_not_exist behavior for missing tasks.
• 2026-03-18: Added Ali Bailian video docs in video/alibailian.mdx and en/video/alibailian.mdx, documenting /alibailian/api/v1/services/aigc/video-generation/video-synthesis and /alibailian/api/v1/tasks/{task_id} with request/response examples and verified local route behavior.
• 2026-03-18: Updated docs.json navigation to surface the new operational health, Claude token counting, Gemini-compatible model-list, video alias, and Ali Bailian pages in both Chinese and English.
• 2026-03-18: With product-direction clarification that https://crazyrouter.com/status is the user-facing surface, removed the public docs pages for GET /api/status/page, GET /api/status/health, GET /healthz/live, and GET /healthz/ready; these routes are now treated as internal page-data / ops / probe endpoints rather than public user documentation.
• 2026-03-18: Added a public-facing https://crazyrouter.com/status link to introduction.mdx and en/introduction.mdx so users are sent to the real service-status surface instead of internal status APIs.
• 2026-03-18: Removed token-management/account.mdx and en/token-management/account.mdx from public docs because GET /api/user/self is a dashboard/internal user endpoint with role, referral, and settings payloads, not a main public integration API.
• 2026-03-18: Renamed the token section in docs.json to API Key & 控制台管理 / API Key & Console Management to clarify that this area is for API-key scope checks and console-side token automation, not the main model-calling API surface.
• 2026-03-18: Added explicit dashboard-auth notes to token-management/create|list|update|delete|search.mdx and their English counterparts, and filled in missing auth sections on the update/delete/search pages so users do not mistake these console APIs for normal sk-xxx model endpoints.
• 2026-03-18: Deepened integrations/claude-code.mdx and en/integrations/claude-code.mdx from quick setup pages into fuller operational guides covering Git installation, Node installation, Claude Code installation, temporary and persistent env vars, repo snapshotting, and staged first-run validation.
• 2026-03-18: Further refined integrations/claude-code.mdx and en/integrations/claude-code.mdx with explicit Windows vs macOS install paths, OS-specific claude verification commands like where.exe / which, clearer temporary-variable echo checks, and PowerShell-vs-Git-Bash guidance for first-time Windows setup.
• 2026-03-18: Deepened integrations/codex.mdx and en/integrations/codex.mdx with full Git / Node prerequisites, OPENAI_API_KEY persistence, exact ~/.codex/config.toml writing steps, repo snapshot guidance, and safer first-run validation order.
• 2026-03-18: Further refined integrations/codex.mdx and en/integrations/codex.mdx with explicit Windows vs macOS install paths, OS-specific codex installation commands, verification commands like where.exe / which, and config-file inspection steps after writing ~/.codex/config.toml.
• 2026-03-18: Deepened integrations/aider.mdx and en/integrations/aider.mdx with Git / Python prerequisites, the official Aider installer flow, env-var persistence, optional ~/.aider.conf.yml, repo snapshot guidance, and staged first validation.
• 2026-03-18: Deepened integrations/cline.mdx and en/integrations/cline.mdx with Git setup, VS Code installation, Cline extension installation, repo snapshot guidance, and a safer read-only-first validation sequence.
• 2026-03-18: Further refined integrations/aider.mdx and en/integrations/aider.mdx with explicit Windows vs macOS install paths, where.exe / which verification, safer persistent env-var checks, and Crazyrouter raw model ID naming for the OpenAI-compatible route.
• 2026-03-18: Further refined integrations/cline.mdx and en/integrations/cline.mdx with explicit Windows vs macOS install paths, optional code command verification, clearer UI-only provider setup guidance, and a note that Cline reuses VS Code proxy settings on restricted networks.
• 2026-03-18: Expanded integrations/langchain.mdx and en/integrations/langchain.mdx into fuller framework guides covering Python and JavaScript package installation, persistent OPENAI_API_KEY handling, minimal chat validation, embeddings, prompt chains, RAG rollout order, and LangChain-specific troubleshooting for Crazyrouter.
• 2026-03-18: Expanded integrations/immersive-translate.mdx and en/integrations/immersive-translate.mdx with the official full-path custom OpenAI URL pattern, safer first-pass model choices, request-rate cautions, and a clearer small-text-first validation flow for Crazyrouter-based translation.
• 2026-03-18: Expanded integrations/coze.mdx and en/integrations/coze.mdx to reposition the public guide around Coze API plugins and workflow HTTP request nodes, explicitly avoid overclaiming enterprise custom-model support, and add staged validation, token-scope guidance, and troubleshooting.
• 2026-03-18: Expanded integrations/chatbox.mdx and en/integrations/chatbox.mdx into fuller desktop guides covering provider selection, root-host plus default-path setup, staged model rollout, knowledge-base rollout order, token hygiene, and practical ChatBox troubleshooting for Crazyrouter.
• 2026-03-18: Expanded integrations/zotero.mdx and en/integrations/zotero.mdx into fuller research-workflow guides that distinguish between plugins with custom OpenAI-compatible endpoint support and plugins that only accept official OpenAI keys, while adding Zotero desktop install notes, staged single-paper validation, token hygiene, and troubleshooting.
• 2026-03-22: Added top-level 核心能力 / Features groups to docs.json so users can find tool calling, structured outputs, web search, reasoning, vision, and PDF capabilities directly from navigation before picking a protocol-specific guide.
• 2026-03-22: Added unified capability hub pages in features/ and en/features/ for tool calling, structured outputs, web search, and reasoning so users can start from capability first, then drill into protocol-specific docs.
• 2026-03-22: Revalidated those four capability pages with real requests against https://crazyrouter.com and tightened them to the current production behavior: gpt-5.4 reasoning now uses Responses reasoning items as the verified path, Claude tool-calling examples now use a stronger reproducible /v1/messages pattern, Claude web search is documented with an explicit prompt that reliably surfaces remote_web_search, and Claude structured output remains excluded from strict-success claims because the response shape is not yet stable.
• 2026-03-22: Reworked chat/openai/reasoning.mdx, chat/responses/gpt5-thinking.mdx, and reference/thinking.mdx plus their English counterparts using the same production recheck: Chat Completions reasoning is now documented as a final-answer path rather than a stable visible-trace path, GPT Responses reasoning is documented with the current reasoning item shape and verified SSE event names, and the thinking reference now distinguishes Responses reasoning, Claude thinking, and Gemini thoughtsTokenCount instead of mixing them together.
• 2026-03-22: Reworked chat/openai/web-search.mdx, chat/anthropic/messages.mdx, and chat/gemini/native.mdx plus their English counterparts using the same production-first rule: OpenAI web search now documents the verified gpt-5.4 Responses path instead of the older Chat web_search pattern, Claude Messages now uses the current claude-sonnet-4-6 / claude-opus-4-6-thinking models and verified auth/tool/thinking behavior, and Gemini Native now uses gemini-3-pro-preview with verified text, SSE streaming, structured outputs, Google Search, and thinking markers.
• 2026-03-22: Reworked chat/gemini/openai-compat.mdx, chat/responses/overview.mdx, and chat/openai/overview.mdx plus their English counterparts so the overview layer no longer lags behind the feature pages: Gemini OpenAI-compatible docs now use the verified gemini-3-pro-preview compatibility path, the Responses overview now frames /v1/responses around the currently verified gpt-5.4 reasoning and web-search behavior, and the Chat Completion object page now treats reasoning_content as optional rather than a guaranteed cross-model field.
• 2026-03-23: Reworked chat/openai/models.mdx, chat/gemini/openai-models.mdx, and reference/tutorial.mdx plus their English counterparts using a fresh same-day production recheck: /v1/models and /v1beta/openai/models are now documented with current response shape, auth, and example model IDs such as gpt-5.4, claude-sonnet-4-6, and gemini-3-pro-preview, while the tutorial now focuses on production-verified starter routes instead of broad generic compatibility claims.
• 2026-03-23: Reworked chat/openai/special-models.mdx, chat/openai/completions.mdx, and chat/anthropic/overview.mdx plus their English counterparts using the same-day production check: the special-models page now documents current availability instead of stale success examples because qwen-mt-turbo was not present and deepseek-ocr was temporarily unavailable, the Chat Completions page now uses the current gpt-5.4 response and SSE chunk shape, and the Anthropic overview now uses claude-sonnet-4-6 with the currently observed native SSE event sequence.
• 2026-03-23: Reworked chat/gemini/tools.mdx, chat/gemini/document.mdx, chat/openai/image-creation.mdx, chat/openai/completions-legacy.mdx, chat/gemini/image-gen.mdx, and chat/gemini/image-edit.mdx plus their English counterparts using same-day production checks: Gemini tools are now documented around the verified gemini-3-pro-preview tool markers, Gemini document understanding is narrowed to verified PDF and schema-output paths, Chat image creation now documents the currently observed text-wrapped gpt-4o-image behavior and points stable automation to gpt-image-1, the legacy completions page now records the current gpt-3.5-turbo-instruct unavailability result, Gemini image generation now explains the observed per-model output-shape differences, and Gemini image editing is narrowed to a current availability note because all same-day retests failed or timed out.
• 2026-03-23: Reworked chat/responses/create.mdx, chat/responses/function-calling.mdx, chat/responses/web-search.mdx, chat/openai/structured-output.mdx, chat/openai/vision.mdx, chat/anthropic/vision.mdx, chat/anthropic/pdf.mdx, chat/anthropic/completions.mdx, and chat/gemini/vision.mdx plus their English counterparts using same-day production checks: Responses docs now use gpt-5.4 and current output markers such as message, function_call, and web_search_call; OpenAI structured output and vision now keep only the revalidated json_schema and data-URL image-input paths; Claude OpenAI-compatible chat, native vision, and native PDF docs now use claude-sonnet-4-6 with current response shapes; and Gemini vision is narrowed to the revalidated gemini-3-pro-preview image-input path.
• 2026-03-23: Reworked chat/openai/function-calling.mdx and en/chat/openai/function-calling.mdx using the same-day production check so the Chat Completions tool-calling page now uses gpt-5.4 and the currently observed tool_calls response shape instead of older gpt-4o examples.
• 2026-03-23: Updated quickstart.mdx, making-requests.mdx, and authentication.mdx plus their English counterparts so the highest-visibility starter examples now use the same-day production-verified gpt-5.4 model instead of older gpt-4o examples.