Anthropic API (Claude)

Environment variables

Claude Code and the Anthropic SDKs read these automatically:

export ANTHROPIC_BASE_URL="https://apixly.ai"
export ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"

Messages API

curl https://apixly.ai/v1/messages \
  -H "x-api-key: YOUR_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

{
  "id": "msg_01ABC...",
  "type": "message",
  "role": "assistant",
  "content": [{ "type": "text", "text": "Hello! How can I help?" }],
  "model": "claude-sonnet-4-6",
  "stop_reason": "end_turn",
  "usage": { "input_tokens": 10, "output_tokens": 12 }
}

Auth uses the x-api-key header plus anthropic-version — not a Bearer token.

Token count

POST /v1/messages/count_tokens accepts model, messages and optional system/tools with the Messages shape. Availability depends on the upstream account; model listing does not guarantee this endpoint. Use the final generation usage for billing reconciliation.

Model-specific request contracts

Discover available IDs with GET /v1/models using your key. The public price catalog is not a guarantee of current routing. Existing valid requests need no changes; optional fields are model-specific.

POST /v1/messages: model,max_tokens,messages; system is top-level. Opus4.5 supports effort=low/medium/high with or without manual enabled thinking; it does not support adaptive. Sonnet/Haiku4.5 do not support effort. Opus4.6/Sonnet4.6 support adaptive and still accept enabled+budget_tokens; budget must be below max_tokens. Opus4.7/4.8 and Claude5 use adaptive instead of enabled+budget; the gateway converts the old shape. xhigh is retained for supported Opus4.7/4.8 and the listed Claude5 models; it is not universal for4.6. Opus5.5 cannot disable thinking; Sonnet5.5 uses between_tools (low/medium/high) to omit up-front thinking, not disabled. SDK base URL is the bare domain.