# claude-sonnet-5 API [Playground](/models/anthropic/claude-sonnet-5) ยท [API reference](/docs/models/anthropic/claude-sonnet-5) ## Authentication and model selection Use your Topview Platform API key, not a provider key. Send Authorization: Bearer YOUR_PLATFORM_API_KEY. Claude also accepts x-api-key. GET https://platform-backend.topview.ai/v1/models lists enabled models. Model-restricted keys must allow anthropic/claude-sonnet-5. Select the model with model: "claude-sonnet-5". Keep API keys on your server. ``` curl https://platform-backend.topview.ai/v1/models -H 'Authorization: Bearer YOUR_PLATFORM_API_KEY' ``` ## Request POST /v1/messages. Text generation returns the response directly, or streams SSE; it does not use the image/video prediction submission and polling API. ``` curl https://platform-backend.topview.ai/v1/messages \ -H 'Authorization: Bearer YOUR_PLATFORM_API_KEY' \ -H 'Content-Type: application/json' \ -H 'Idempotency-Key: unique_request_0001' \ -d '{ "model": "claude-sonnet-5", "messages": [ { "role": "user", "content": "Hello" } ], "max_tokens": 1024 }' ``` ## Parameters and limits Required: model and messages (1โ€“64 messages). max_tokens: integer 1โ€“8192, default 1024. stream: boolean, default false. Maximum request JSON: 32 MiB; attachments total at most 20 MiB. Send the conversation history with every request. Use the top-level system field for system instructions. Optional fields: temperature, top_p, stop_sequences, thinking, output_config and cache_control. Model-specific parameter compatibility still applies. tools, tool_choice and metadata are accepted; at most 16 client-side function tools. Hosted tools, premium processing tiers and stored server-side conversations are not enabled. ## Response and token usage Read text blocks in content[].text and usage.input_tokens / usage.output_tokens. Tool calls appear as tool_use blocks. Cache usage may include cache_read_input_tokens and cache_creation_input_tokens. ``` { "type": "message", "role": "assistant", "model": "claude-sonnet-5", "content": [ { "type": "text", "text": "Hello!" } ], "stop_reason": "end_turn", "usage": { "input_tokens": 8, "output_tokens": 3 } } ``` ## Streaming and tools Set stream: true and consume text/event-stream incrementally. Handle message_start, content_block_start, content_block_delta, message_delta and message_stop. Assemble tool input JSON deltas before running the tool. Function tools run in your application: execute the requested function, append the tool result and send the next turn. A stream disconnect does not cancel the provider call or imply zero charge. ## Python Install requests. Use a new idempotency key for a new logical request; retain it for retries of that same request. ``` import os, uuid, requests response = requests.post( "https://platform-backend.topview.ai/v1/messages", headers={"Authorization": "Bearer " + os.environ["TOPVIEW_API_KEY"], "Idempotency-Key": uuid.uuid4().hex}, json={"model":"claude-sonnet-5","messages":[{"role":"user","content":"Hello"}],"max_tokens":1024}, timeout=180, ) print("Request ID:", response.headers.get("X-Platform-Request-Id")) response.raise_for_status() print(response.json()) ``` ## JavaScript Run on your server. Inspect non-2xx responses before parsing generated output. ``` const response = await fetch("https://platform-backend.topview.ai/v1/messages", { method: "POST", headers: {Authorization: "Bearer " + process.env.TOPVIEW_API_KEY, "Content-Type": "application/json", "Idempotency-Key": crypto.randomUUID()}, body: JSON.stringify({ "model": "claude-sonnet-5", "messages": [ { "role": "user", "content": "Hello" } ], "max_tokens": 1024 }), }); const data = await response.json(); if (!response.ok) throw new Error(JSON.stringify(data)); console.log(response.headers.get("X-Platform-Request-Id"), data); ``` ## Attachments and URLs Messages accepts PNG/JPEG/WebP/non-animated GIF image blocks and PDF document blocks with Base64 or public HTTPS URL sources. Plain-text document sources and content-block document sources are supported. Use source.type=text with media_type=text/plain for extracted text. Native Word/Excel files are not accepted by this Messages route; convert them to PDF or extract their text. Playground reads text files and CSV as text. ``` { "model": "claude-sonnet-5", "max_tokens": 1024, "messages": [ { "role": "user", "content": [ { "type": "document", "source": { "type": "url", "url": "https://example.com/document.pdf" } }, { "type": "image", "source": { "type": "url", "url": "https://example.com/image.png" } }, { "type": "text", "text": "Summarize the document and image." } ] } ] } ``` ## Attachment delivery and limits Platform limit: 20 MiB combined decoded attachments, including generated GPT PDF page images, and 32 MiB request JSON. Up to 600 images / PDF pages, subject to each model context and upstream limits. URLs must be publicly readable HTTPS direct file links without credentials or redirects; private networks are rejected. URLs are fetched and frozen before reservation. Base64 image sources use the matching media type; PDF uses application/pdf. GPT file_data accepts a data URL (or raw Base64 with a .pdf filename). Provider file_id/Files APIs are not exposed by this supplier. A web page URL in plain text is only text; hosted web-fetch/search tools are not enabled. Image and PDF page tokens contribute to input usage; an attachment reservation is conservative and unused funds are released. The API response retains PDF citations when requested through the native protocol. ## Pricing and billing Prices are USD per million tokens: input 1.7; output 8.5; cache read 0.17. Cache write (5 minutes): 2.125; cache write (1 hour): 3.4. The wallet reserves a conservative budget before submission, charges actual reported token usage and releases the unused reservation. The reservation is not the final charge. Check Usage for settlement and the request ID. Image/video promotions should not be assumed to apply to text. ## Errors, retries and idempotency Save X-Platform-Request-Id. Supply Idempotency-Key (16โ€“128 letters, digits, underscore or hyphen). Reuse the same key and identical payload with the same API credential when recovering the same request. A settled replay does not call the provider again; an active or uncertain request returns 409. A new key creates a new paid request. Invalid input returns 400; authentication or access errors use 401/403. Do not blindly resubmit after a timeout, 5xx or interrupted stream: provider execution may have occurred. Check Usage and the original request ID; uncertain usage requires reconciliation. A confirmed provider rejection releases the reservation.