generated: '2026-08-27'
method: searched
source: https://platform.claude.com/playground, https://platform.claude.com/docs/en/release-notes/overview
docs: https://support.claude.com/en/articles/8606378-how-do-i-use-playground
notes: >-
  Anthropic publishes an interactive CONSOLE PLAYGROUND, not a test-mode API.
  This distinction matters for anyone budgeting an integration: there is no
  sandbox key, no test-vs-live key prefix, no free simulated traffic and no
  fixture/trigger tooling. Every call you make while developing is a real,
  billed inference call against production. A payments-style test mode has no
  equivalent here.
console:
  name: Playground
  url: https://platform.claude.com/playground
  former_name: Workbench
  renamed: '2026-08-18'
  renamed_note: >-
    Workbench was renamed to Playground in the Claude Console on 2026-08-18.
    https://platform.claude.com/workbench still resolved (HTTP 200) when probed
    2026-08-27, but /playground is the current canonical path.
  requires_login: true
  capabilities:
  - Supports every Messages API parameter.
  - Ships templates demonstrating API features such as code execution and web search.
  - Shows the full SDK request and the API response for each run.
test_mode:
  supported: false
  test_key_prefix: null
  live_key_prefix: sk-ant-
  admin_key_prefix: sk-ant-admin
  webhook_signing_secret_prefix: whsec_
  note: >-
    Key prefixes distinguish ROLE (standard vs Admin API), not ENVIRONMENT. There
    is no test-mode credential. Isolate development traffic with a separate
    workspace and its own key, and cap it with the workspace spend limit — that is
    the closest available equivalent to a sandbox.
test_fixtures:
  test_values_published: false
  time_simulation: false
  trigger_tooling: false
  note: Anthropic publishes no magic test identifiers, test cards, test clocks or
    event-trigger tooling. Do not invent any.
free_evaluation:
  free_tier: false
  note: >-
    Access is credit-based. Tier 1 begins at a $5 credit purchase (see
    plans/anthropic-plans-pricing.yml and rate-limits/anthropic-rate-limits.yml).
    Token counting (POST /v1/messages/count_tokens, operationId countMessageTokens)
    is the one way to size a request before paying for it — it is the practical
    dry-run primitive on this API.
dry_run:
  supported: partial
  mechanism: countMessageTokens (POST /v1/messages/count_tokens)
  note: >-
    Counts tokens for a fully-formed Messages request, including tools and system
    prompt, without creating the message or billing for inference. It rehearses
    the COST of a call, not its effect. No write operation elsewhere on the API
    accepts a dry-run flag.
related_environments:
  managed_agents_sandboxes: https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes
  note: >-
    Managed Agents "sandboxes" are execution environments for agent code — where
    an agent's tools run — not an API test environment. Do not conflate the two.