{
  "title": "Prompt caching",
  "description": "Reuse eligible repeated prompt content across requests under documented caching behavior.",
  "slug": "prompt-cache-reuse",
  "locale": "en",
  "seoTitle": "Prompt and context caching compatibility — Can My Agent Use",
  "socialTitle": "Prompt caching",
  "socialDescription": "Compare automatic and persistent reuse of eligible repeated context across harnesses.",
  "llmSummary": "Prompt caching reuses eligible repeated prompt content under documented latency or billing behavior. Browser, HTTP, file, and retrieval caching do not count.",
  "audience": "Engineers comparing latency and token cost for repeated large contexts.",
  "contentKind": "feature",
  "status": "published",
  "tags": [
    "context",
    "prompt-caching",
    "latency",
    "cost"
  ],
  "updated": "2026-08-28T00:00:00.000Z",
  "published": "2026-08-28T00:00:00.000Z",
  "category": "models-context",
  "summary": "Reuse eligible repeated prompt content under documented caching behavior.",
  "specLabel": "Common product term",
  "aliases": [
    "prompt caching",
    "context caching",
    "cached input",
    "prefix caching"
  ],
  "capabilityKind": "atomic",
  "parent": "models-and-context",
  "related": [
    "prompt-cache-controls",
    "prompt-cache-telemetry",
    "usage-metering"
  ],
  "relations": [],
  "highlight": true,
  "notes": [
    {
      "id": 1,
      "text": "Evidence checked 2026-08-28: Claude Code documents automatic prefix-based prompt caching for its system prompt, project context, conversation history, and tool results, with explicit invalidation behavior."
    },
    {
      "id": 2,
      "text": "Evidence checked 2026-08-28: Gemini CLI documents automatic token caching for Gemini API-key and Vertex AI authentication, while OAuth through Code Assist does not support cached-content creation."
    }
  ],
  "issues": [],
  "resources": [
    {
      "title": "Methodology",
      "href": "/methodology",
      "kind": "note"
    },
    {
      "id": "anthropic-code-prompt-caching",
      "title": "Anthropic — How Claude Code uses prompt caching",
      "href": "https://code.claude.com/docs/en/prompt-caching",
      "kind": "docs",
      "publisher": "Anthropic",
      "evidenceType": "documented",
      "reviewedAt": "2026-08-28"
    },
    {
      "id": "google-gemini-cli-token-caching",
      "title": "Google — Gemini CLI token caching",
      "href": "https://geminicli.com/docs/cli/token-caching/",
      "kind": "docs",
      "publisher": "Google",
      "evidenceType": "documented",
      "reviewedAt": "2026-08-28"
    }
  ],
  "support": [
    {
      "harness": "claude-cli",
      "versions": [
        {
          "track": "current",
          "status": "yes",
          "noteIds": [
            1
          ],
          "target": {
            "kind": "dated-documentation",
            "revision": "current Claude Code documentation",
            "observedAt": "2026-08-28"
          },
          "environmentProfile": "local-default",
          "qualifiers": [
            {
              "type": "runtime",
              "value": "automatic exact-prefix caching covers stable request layers; switching models, reconnecting MCP servers, compaction, and upgrades can invalidate all or part of the prefix"
            },
            {
              "type": "policy",
              "value": "cache infrastructure and retention depend on the authentication and serving provider"
            }
          ],
          "evidence": [
            {
              "resourceId": "anthropic-code-prompt-caching",
              "type": "documented",
              "observedAt": "2026-08-28"
            }
          ]
        }
      ]
    },
    {
      "harness": "gemini-cli",
      "versions": [
        {
          "track": "current",
          "status": "partial",
          "noteIds": [
            2
          ],
          "target": {
            "kind": "dated-documentation",
            "revision": "current Gemini CLI documentation",
            "observedAt": "2026-08-28"
          },
          "environmentProfile": "local-default",
          "qualifiers": [
            {
              "type": "runtime",
              "value": "automatic caching reuses previous system instructions and context for Gemini API-key and Vertex AI users"
            },
            {
              "type": "policy",
              "value": "OAuth users through Google Personal or Enterprise Code Assist do not receive cached-content creation"
            }
          ],
          "evidence": [
            {
              "resourceId": "google-gemini-cli-token-caching",
              "type": "documented",
              "observedAt": "2026-08-28"
            }
          ]
        }
      ]
    }
  ],
  "html": "/features/prompt-cache-reuse",
  "markdown": "/features/prompt-cache-reuse.md",
  "json": "/api/v1/features/prompt-cache-reuse.json",
  "body": "This row concerns model-request prompt or context caching: eligible repeated prefixes receive documented processing, latency, or billing reuse. It does not include browser caches, downloaded-file caches, embedding indexes, retrieval caches, build caches, or a conversation merely retaining its history.\n\nEvidence should identify whether reuse is automatic or explicit, minimum eligible prefix size, exact-prefix requirements, supported models and regions, cache lifetime, isolation boundary, invalidation behavior, and read-versus-write pricing. A model API feature does not prove that a hosted chat or coding harness preserves stable prefixes or passes cache controls through."
}