Skip to content

Providers / Amazon Bedrock · All access methods for openai/gpt-6-luna

GPT-6 Luna (Global)

amazon-bedrock/global.openai.gpt-6-luna API key models.dev2026-11-17providers.dev2026-11-17

Type chat · Released 2026-09-22

Input window
922,000 tokensdocumented
models.dev · 2026-11-17
Input alone
Context window
1,050,000 tokens
Exposed by this provider
Maximum output
128,000 tokens
Per response; not a default
Cache lifetime, short
30 mindocumented
providers.dev · 2026-11-17
Cache lifetime, long
Unknown
Not documented or measured

Limits

Input window

LimitValueSource
Input window
input_window: the most input one request can carry on this entry, in tokens as the provider counts them.
922,000 tokensdocumentedInput alonemodels.dev2026-11-17
Documented
The figure the source documents, and the field it is in.
922,000 tokenslimit.inputmodels.dev2026-11-17
Measured
What each limit evaluation that counts gives, one line per evaluation.
Not measured–

How the input window is worked out, and what it does not say

Documented

LimitValueSource
Context window
limit.context: what this provider accepts for the model, input and output together.
1,050,000 tokensmodels.dev
Largest input
limit.input: stated where the input limit is lower than the context window.
922,000 tokensmodels.dev
Maximum output
limit.output: the most one response can hold. A maximum, not what you get by default.
128,000 tokensmodels.dev

Measured tokens

DateMeasured throughContextInput acceptedInput refusedOutput limitRequested output acceptedEvidence
No limits were measured on this entry.

Price

Base chargeRate
Input$0.10 / 1M tokens
Output$0.50 / 1M tokens
Reasoning outputUnknown
Cache read$0.01 / 1M tokens
Cache write$0.125 / 1M tokens · 30 min lifetime

Conditional rates

ConditionChargeRateApplies to
Context above 272,000 tokensInput$0.20 / 1M tokensWhole request
Output$0.75 / 1M tokens
Cache read$0.02 / 1M tokens
Context above 272,000 tokens and cache lifetime 30 minCache write$0.25 / 1M tokensWhole request

Source: models.dev 2026-11-17

Other prices in the source, not shown above (0)

None

Request controls

ControlSupportRequest fieldAccepted valuesDefaultConditions
Service tierUnknownUnknownUnknownUnknownUnknown

Source: Unknown

Prompt caching

Lifetimes

Every cache lifetime of this entry, by retention
LifetimeDocumented for, or measured throughTime since the prompt was cached · served from the cache, in percent, with hits/checks
At once2.5 min5 min7.5 min10 min15 min20 min25 min30 min40 min50 min1 h1 h 151 h 302 h4 h8 h12 h24 h
Short retention
Safe lifetime
30 mindocumentedproviders.dev2026-11-17
Documented
30 min
Measured
Not measured
Inside the documented lifetime
Not applicable
30 mindocumentedproviders.dev2026-11-17Manual explicit
As documented: 30m
Not applicable
Long retention
Safe lifetime
Unknown
Documented
Not documented
Measured
Not measured
Inside the documented lifetime
Not applicable

Checks served from the cache, in percent and as hits/checks, by time since the prompt was cached100 All67 Half or more33 Fewer0 None· Not checked at that ageMethodology

Cache modes

ModeSupportedMin. tokensBreakpointsTargets
Automatic implicit Supported 1,024 Not applicable Not applicable
Manual explicit Supported Unknown Unknown Unknown

Source: providers.dev · Cache modes

Retention

ControlSupportRequest fieldAccepted valuesDefaultConditions
Cache retentionSupportedprompt_cache_options.ttl
Request body, top level
Short: 30m
30m30m is the only supported value.
Explicit caching also needs prompt_cache_options.mode set to explicit and a prompt_cache_breakpoint on a content block. Converse has no explicit caching for these models; GPT-6.1 Terra has it in the Responses API only.

Source: docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html 2026-11-17 · docs.aws.amazon.com…userguide/model-card-openai-gpt-6-1-sol.html 2026-11-17 · docs.aws.amazon.com…erguide/model-card-openai-gpt-6-1-terra.html 2026-11-17

Provenance

ProvenanceValue
Sourcemodels.dev2026-11-17providers.dev2026-11-17
Inherited fromNone
Corrections

capabilities.cache.explicit.ttl is ["30m"]. models.dev says: nothing.

The Bedrock user guide documents a 30-minute lifetime for explicit prompt caching on GPT-5.6 and later models, set with prompt_cache_options.ttl, and says prompt caching can be used with cross-Region inference. These entries are the global. and us. inference profiles of those models.

  • documented docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html 2026-11-17
    Supported models table: 'GPT-6.1 Sol | openai.gpt-6.1-sol | Generally Available | 1,024 | 4 cache writes | 30 minutes', the same Supported TTL for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. GPT-5.6 and later models ('later models such as GPT-6 Astra (openai.gpt-6-astra) and GPT-6.1 Sol (openai.gpt-6.1-sol)'): '30-minute minimum TTL — Cached prefixes remain available for reuse for at least 30 minutes [...] The TTL is set via prompt_cache_options.ttl, and 30m is the only supported value.' 'The cache has a Time To Live (TTL), which resets with each successful cache hit.' Prompt Caching with Cross-region Inference: 'Prompt caching can be used in conjunction with cross region inference.'
  • documented docs.aws.amazon.com…userguide/model-card-openai-gpt-6-1-sol.html 2026-11-17
    'For explicit caching, set prompt_cache_options.mode to explicit. Set prompt_cache_options.ttl to 30m, the only supported TTL and the default.' (the same sentence on the cards of GPT-6 Sol and GPT-6 Luna). 'Use us.openai.gpt-6.1-sol for US geographic cross-Region inference or global.openai.gpt-6.1-sol for global cross-Region inference. Direct in-Region invocation is not supported on bedrock-runtime.'
  • documented docs.aws.amazon.com…erguide/model-card-openai-gpt-6-1-terra.html 2026-11-17
    GPT-6.1 Terra is not in the guide's table. Its card: 'Explicit Prompt Caching (Responses API only)'. 'The 30-minute figure is how long a cached prompt prefix is retained, not a quantity of tokens.' The 30-minute cache write is priced for Geo CRIS and Global CRIS; bedrock-runtime 'requests must name the geographic cross-Region inference ID us.openai.gpt-6.1-terra'.
Evidence
  • documented docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html 2026-11-17
    Supported models table: 'GPT-6.1 Sol | openai.gpt-6.1-sol | Generally Available | 1,024 | 4 cache writes | 30 minutes', the same Supported TTL for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. GPT-5.6 and later models ('later models such as GPT-6 Astra (openai.gpt-6-astra) and GPT-6.1 Sol (openai.gpt-6.1-sol)'): '30-minute minimum TTL — Cached prefixes remain available for reuse for at least 30 minutes [...] The TTL is set via prompt_cache_options.ttl, and 30m is the only supported value.' 'The cache has a Time To Live (TTL), which resets with each successful cache hit.' Prompt Caching with Cross-region Inference: 'Prompt caching can be used in conjunction with cross region inference.'
  • documented docs.aws.amazon.com…userguide/model-card-openai-gpt-6-1-sol.html 2026-11-17
    'For explicit caching, set prompt_cache_options.mode to explicit. Set prompt_cache_options.ttl to 30m, the only supported TTL and the default.' (the same sentence on the cards of GPT-6 Sol and GPT-6 Luna). 'Use us.openai.gpt-6.1-sol for US geographic cross-Region inference or global.openai.gpt-6.1-sol for global cross-Region inference. Direct in-Region invocation is not supported on bedrock-runtime.'
  • documented docs.aws.amazon.com…erguide/model-card-openai-gpt-6-1-terra.html 2026-11-17
    GPT-6.1 Terra is not in the guide's table. Its card: 'Explicit Prompt Caching (Responses API only)'. 'The 30-minute figure is how long a cached prompt prefix is retained, not a quantity of tokens.' The 30-minute cache write is priced for Geo CRIS and Global CRIS; bedrock-runtime 'requests must name the geographic cross-Region inference ID us.openai.gpt-6.1-terra'.

Data

GET /api/v1/entries/amazon-bedrock/global.openai.gpt-6-luna

Show the full entry as published
{
  "id": "global.openai.gpt-6-luna",
  "name": "GPT-6 Luna (Global)",
  "description": "OpenAI's most efficient model for focused, high-volume tasks",
  "type": "chat",
  "family": "gpt-luna",
  "canonical_id": "openai/gpt-6-luna",
  "release_date": "2026-09-22",
  "limit": {
    "context": 1050000,
    "input": 922000,
    "output": 128000
  },
  "capabilities": {
    "cache": {
      "implicit": {
        "min_tokens": 1024
      },
      "explicit": {
        "ttl": [
          "30m"
        ]
      }
    },
    "temperature": true,
    "top_k": false,
    "top_p": true,
    "structured_output": true,
    "system_updates": false,
    "effort_updates": false,
    "tools": {
      "supported": true
    },
    "reasoning": {
      "supported": true,
      "effort": [
        "none",
        "low",
        "medium",
        "high",
        "xhigh",
        "max"
      ]
    }
  },
  "cost": {
    "text": {
      "input": {
        "price": 0.1,
        "cache": {
          "read": 0.01,
          "write": {
            "30m": 0.125
          }
        },
        "overrides": [
          {
            "when": {
              "context": {
                "gt": 272000
              }
            },
            "price": 0.2,
            "cache": {
              "read": 0.02,
              "write": {
                "30m": 0.25
              }
            }
          }
        ]
      },
      "output": {
        "price": 0.5,
        "overrides": [
          {
            "when": {
              "context": {
                "gt": 272000
              }
            },
            "price": 0.75
          }
        ]
      }
    }
  },
  "modalities": {
    "input": [
      "text",
      "image/jpeg",
      "image/png",
      "image/gif",
      "image/webp"
    ],
    "output": [
      "text"
    ]
  },
  "api": {
    "converse": {
      "base_url": "https://bedrock-runtime.${AWS_REGION}.amazonaws.com",
      "sdk": {
        "aisdk": "@ai-sdk/amazon-bedrock",
        "opencodeai": "@opencode/ai/providers/amazon-bedrock"
      },
      "capabilities": {
        "tools": {
          "updates": []
        }
      }
    }
  },
  "open_weights": false,
  "knowledge": "2026-05-18",
  "last_updated": "2026-09-22",
  "providers_dev": {
    "origin": "upstream",
    "access": "api",
    "provenance": [
      {
        "path": "capabilities.cache.explicit.ttl",
        "op": "set",
        "value": [
          "30m"
        ],
        "had_upstream": false,
        "upstream": null,
        "reason": "The Bedrock user guide documents a 30-minute lifetime for explicit prompt caching on GPT-5.6 and later models, set with prompt_cache_options.ttl, and says prompt caching can be used with cross-Region inference. These entries are the global. and us. inference profiles of those models.",
        "evidence": [
          {
            "kind": "documented",
            "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html",
            "date": "2026-11-17",
            "note": "Supported models table: 'GPT-6.1 Sol | openai.gpt-6.1-sol | Generally Available | 1,024 | 4 cache writes | 30 minutes', the same Supported TTL for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. GPT-5.6 and later models ('later models such as GPT-6 Astra (openai.gpt-6-astra) and GPT-6.1 Sol (openai.gpt-6.1-sol)'): '30-minute minimum TTL — Cached prefixes remain available for reuse for at least 30 minutes [...] The TTL is set via prompt_cache_options.ttl, and 30m is the only supported value.' 'The cache has a Time To Live (TTL), which resets with each successful cache hit.' Prompt Caching with Cross-region Inference: 'Prompt caching can be used in conjunction with cross region inference.'"
          },
          {
            "kind": "documented",
            "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html",
            "date": "2026-11-17",
            "note": "'For explicit caching, set prompt_cache_options.mode to explicit. Set prompt_cache_options.ttl to 30m, the only supported TTL and the default.' (the same sentence on the cards of GPT-6 Sol and GPT-6 Luna). 'Use us.openai.gpt-6.1-sol for US geographic cross-Region inference or global.openai.gpt-6.1-sol for global cross-Region inference. Direct in-Region invocation is not supported on bedrock-runtime.'"
          },
          {
            "kind": "documented",
            "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-terra.html",
            "date": "2026-11-17",
            "note": "GPT-6.1 Terra is not in the guide's table. Its card: 'Explicit Prompt Caching (Responses API only)'. 'The 30-minute figure is how long a cached prompt prefix is retained, not a quantity of tokens.' The 30-minute cache write is priced for Geo CRIS and Global CRIS; bedrock-runtime 'requests must name the geographic cross-Region inference ID us.openai.gpt-6.1-terra'."
          }
        ]
      }
    ],
    "inherited": null,
    "cache": {
      "documented": {
        "support": "supported",
        "modes": {
          "implicit": {
            "support": "supported",
            "min_tokens": 1024,
            "max_breakpoints": null,
            "targets": null
          },
          "explicit": {
            "support": "supported",
            "min_tokens": null,
            "max_breakpoints": null,
            "targets": null
          }
        }
      },
      "retention": {
        "support": "supported",
        "parameter": {
          "name": "prompt_cache_options.ttl",
          "location": "Request body, top level"
        },
        "accepted_values": [
          "30m"
        ],
        "default": "30m",
        "profiles": [
          {
            "profile": "short",
            "value": "30m"
          }
        ],
        "conditions": [
          "30m is the only supported value.",
          "Explicit caching also needs prompt_cache_options.mode set to explicit and a prompt_cache_breakpoint on a content block. Converse has no explicit caching for these models; GPT-6.1 Terra has it in the Responses API only."
        ],
        "evidence": [
          {
            "kind": "documented",
            "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html",
            "date": "2026-11-17",
            "note": "Supported models table: 'GPT-6.1 Sol | openai.gpt-6.1-sol | Generally Available | 1,024 | 4 cache writes | 30 minutes', the same Supported TTL for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. GPT-5.6 and later models ('later models such as GPT-6 Astra (openai.gpt-6-astra) and GPT-6.1 Sol (openai.gpt-6.1-sol)'): '30-minute minimum TTL — Cached prefixes remain available for reuse for at least 30 minutes [...] The TTL is set via prompt_cache_options.ttl, and 30m is the only supported value.' 'The cache has a Time To Live (TTL), which resets with each successful cache hit.' Prompt Caching with Cross-region Inference: 'Prompt caching can be used in conjunction with cross region inference.'"
          },
          {
            "kind": "documented",
            "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html",
            "date": "2026-11-17",
            "note": "'For explicit caching, set prompt_cache_options.mode to explicit. Set prompt_cache_options.ttl to 30m, the only supported TTL and the default.' (the same sentence on the cards of GPT-6 Sol and GPT-6 Luna). 'Use us.openai.gpt-6.1-sol for US geographic cross-Region inference or global.openai.gpt-6.1-sol for global cross-Region inference. Direct in-Region invocation is not supported on bedrock-runtime.'"
          },
          {
            "kind": "documented",
            "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-terra.html",
            "date": "2026-11-17",
            "note": "GPT-6.1 Terra is not in the guide's table. Its card: 'Explicit Prompt Caching (Responses API only)'. 'The 30-minute figure is how long a cached prompt prefix is retained, not a quantity of tokens.' The 30-minute cache write is priced for Geo CRIS and Global CRIS; bedrock-runtime 'requests must name the geographic cross-Region inference ID us.openai.gpt-6.1-terra'."
          }
        ]
      },
      "lifetime": {
        "short": {
          "safe_s": 1800,
          "basis": "documented",
          "source": "providers.dev",
          "date": "2026-11-17",
          "documented_s": 1800,
          "measured_s": null,
          "documented_status": null,
          "clients": 0,
          "evaluations": 0,
          "checks": 0,
          "hits": 0,
          "checks_inside": null,
          "hits_inside": null,
          "tested_from_s": null,
          "first_miss_s": null
        },
        "long": {
          "safe_s": null,
          "basis": "none",
          "source": null,
          "date": null,
          "documented_s": null,
          "measured_s": null,
          "documented_status": null,
          "clients": 0,
          "evaluations": 0,
          "checks": 0,
          "hits": 0,
          "checks_inside": null,
          "hits_inside": null,
          "tested_from_s": null,
          "first_miss_s": null
        },
        "lifetimes": [
          {
            "retention_profile": "short",
            "basis": "documented",
            "source": "providers.dev",
            "date": "2026-11-17",
            "lifetime_s": 1800,
            "mode": "explicit",
            "documented_as": "30m",
            "client": null,
            "client_label": null,
            "parameter": null,
            "value": null,
            "evaluations": 0,
            "evaluation_ids": [],
            "checks": 0,
            "hits": 0,
            "tested_from_s": null,
            "first_miss_s": null,
            "results": null
          }
        ]
      },
      "evaluations": []
    },
    "pricing": {
      "coverage": "partial",
      "billing": "per_use",
      "reference_ref": null,
      "rules": [
        {
          "id": "cost.text.input.price",
          "when": null,
          "meter": "text.input",
          "operation": "set",
          "amount": 0.1,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.input.price",
          "period": null,
          "resulting_rates": null
        },
        {
          "id": "cost.text.input.cache.read",
          "when": null,
          "meter": "text.cache_read",
          "operation": "set",
          "amount": 0.01,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.input.cache.read",
          "period": null,
          "resulting_rates": null
        },
        {
          "id": "cost.text.output.price",
          "when": null,
          "meter": "text.output",
          "operation": "set",
          "amount": 0.5,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.output.price",
          "period": null,
          "resulting_rates": null
        },
        {
          "id": "cost.text.input.cache.write.30m",
          "when": {
            "kind": "predicate",
            "field": "cache.retention_s",
            "op": "eq",
            "value": 1800
          },
          "meter": "text.cache_write",
          "operation": "set",
          "amount": 0.125,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.input.cache.write.30m",
          "period": null,
          "resulting_rates": null
        },
        {
          "id": "cost.text.input.overrides[0].price",
          "when": {
            "kind": "predicate",
            "field": "request.context_tokens",
            "op": "gt",
            "value": 272000
          },
          "meter": "text.input",
          "operation": "set",
          "amount": 0.2,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.input.overrides[0].price",
          "period": null,
          "resulting_rates": null
        },
        {
          "id": "cost.text.input.overrides[0].cache.read",
          "when": {
            "kind": "predicate",
            "field": "request.context_tokens",
            "op": "gt",
            "value": 272000
          },
          "meter": "text.cache_read",
          "operation": "set",
          "amount": 0.02,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.input.overrides[0].cache.read",
          "period": null,
          "resulting_rates": null
        },
        {
          "id": "cost.text.input.overrides[0].cache.write.30m",
          "when": {
            "kind": "all",
            "conditions": [
              {
                "kind": "predicate",
                "field": "request.context_tokens",
                "op": "gt",
                "value": 272000
              },
              {
                "kind": "predicate",
                "field": "cache.retention_s",
                "op": "eq",
                "value": 1800
              }
            ]
          },
          "meter": "text.cache_write",
          "operation": "set",
          "amount": 0.25,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.input.overrides[0].cache.write.30m",
          "period": null,
          "resulting_rates": null
        },
        {
          "id": "cost.text.output.overrides[0].price",
          "when": {
            "kind": "predicate",
            "field": "request.context_tokens",
            "op": "gt",
            "value": 272000
          },
          "meter": "text.output",
          "operation": "set",
          "amount": 0.75,
          "currency": "USD",
          "unit": {
            "name": "token",
            "quantity": 1000000
          },
          "applies_to": "all_usage",
          "source_path": "cost.text.output.overrides[0].price",
          "period": null,
          "resulting_rates": null
        }
      ],
      "rule_order": "listed",
      "unparsed": [],
      "evidence": [
        {
          "kind": "documented",
          "url": "https://models.dev",
          "date": "2026-11-17",
          "note": "The cost models.dev records for this entry. Rules are in the order models.dev applies them: base prices, overrides in their order, then multipliers."
        }
      ]
    },
    "request_controls": {
      "coverage": "partial",
      "items": [
        {
          "id": "cache.retention",
          "support": "supported",
          "when": null,
          "bindings": [
            {
              "location": "body",
              "path": "/prompt_cache_options/ttl",
              "scope": "request"
            }
          ],
          "options": [
            {
              "value": "30m",
              "attributes": {
                "retention_s": 1800,
                "profile": "short"
              },
              "settings": null,
              "when": null
            }
          ],
          "default": {
            "value": "30m"
          },
          "conditions": [
            "30m is the only supported value.",
            "Explicit caching also needs prompt_cache_options.mode set to explicit and a prompt_cache_breakpoint on a content block. Converse has no explicit caching for these models; GPT-6.1 Terra has it in the Responses API only."
          ],
          "evidence": [
            {
              "kind": "documented",
              "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html",
              "date": "2026-11-17",
              "note": "Supported models table: 'GPT-6.1 Sol | openai.gpt-6.1-sol | Generally Available | 1,024 | 4 cache writes | 30 minutes', the same Supported TTL for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. GPT-5.6 and later models ('later models such as GPT-6 Astra (openai.gpt-6-astra) and GPT-6.1 Sol (openai.gpt-6.1-sol)'): '30-minute minimum TTL — Cached prefixes remain available for reuse for at least 30 minutes [...] The TTL is set via prompt_cache_options.ttl, and 30m is the only supported value.' 'The cache has a Time To Live (TTL), which resets with each successful cache hit.' Prompt Caching with Cross-region Inference: 'Prompt caching can be used in conjunction with cross region inference.'"
            },
            {
              "kind": "documented",
              "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html",
              "date": "2026-11-17",
              "note": "'For explicit caching, set prompt_cache_options.mode to explicit. Set prompt_cache_options.ttl to 30m, the only supported TTL and the default.' (the same sentence on the cards of GPT-6 Sol and GPT-6 Luna). 'Use us.openai.gpt-6.1-sol for US geographic cross-Region inference or global.openai.gpt-6.1-sol for global cross-Region inference. Direct in-Region invocation is not supported on bedrock-runtime.'"
            },
            {
              "kind": "documented",
              "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-terra.html",
              "date": "2026-11-17",
              "note": "GPT-6.1 Terra is not in the guide's table. Its card: 'Explicit Prompt Caching (Responses API only)'. 'The 30-minute figure is how long a cached prompt prefix is retained, not a quantity of tokens.' The 30-minute cache write is priced for Geo CRIS and Global CRIS; bedrock-runtime 'requests must name the geographic cross-Region inference ID us.openai.gpt-6.1-terra'."
            }
          ]
        },
        {
          "id": "service_tier",
          "support": "unknown",
          "when": null,
          "bindings": null,
          "options": null,
          "default": null,
          "conditions": [],
          "evidence": []
        }
      ]
    },
    "limits": {
      "measured": [],
      "from_measurement": [],
      "input_window": {
        "tokens": 922000,
        "basis": "documented",
        "bound": "limit",
        "scope": "input",
        "source": "models.dev",
        "date": "2026-11-17",
        "documented_tokens": 922000,
        "documented_field": "limit.input",
        "documented_source": "models.dev",
        "documented_date": "2026-11-17",
        "documented_status": null,
        "measured_tokens": null,
        "measured_bound": null,
        "measured_kind": null,
        "measured_refused": null,
        "measured_date": null,
        "clients": 0,
        "evaluations": 0,
        "measurements": []
      }
    },
    "evidence": [
      {
        "kind": "documented",
        "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/prompt-caching.html",
        "date": "2026-11-17",
        "note": "Supported models table: 'GPT-6.1 Sol | openai.gpt-6.1-sol | Generally Available | 1,024 | 4 cache writes | 30 minutes', the same Supported TTL for GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. GPT-5.6 and later models ('later models such as GPT-6 Astra (openai.gpt-6-astra) and GPT-6.1 Sol (openai.gpt-6.1-sol)'): '30-minute minimum TTL — Cached prefixes remain available for reuse for at least 30 minutes [...] The TTL is set via prompt_cache_options.ttl, and 30m is the only supported value.' 'The cache has a Time To Live (TTL), which resets with each successful cache hit.' Prompt Caching with Cross-region Inference: 'Prompt caching can be used in conjunction with cross region inference.'"
      },
      {
        "kind": "documented",
        "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-sol.html",
        "date": "2026-11-17",
        "note": "'For explicit caching, set prompt_cache_options.mode to explicit. Set prompt_cache_options.ttl to 30m, the only supported TTL and the default.' (the same sentence on the cards of GPT-6 Sol and GPT-6 Luna). 'Use us.openai.gpt-6.1-sol for US geographic cross-Region inference or global.openai.gpt-6.1-sol for global cross-Region inference. Direct in-Region invocation is not supported on bedrock-runtime.'"
      },
      {
        "kind": "documented",
        "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-terra.html",
        "date": "2026-11-17",
        "note": "GPT-6.1 Terra is not in the guide's table. Its card: 'Explicit Prompt Caching (Responses API only)'. 'The 30-minute figure is how long a cached prompt prefix is retained, not a quantity of tokens.' The 30-minute cache write is priced for Geo CRIS and Global CRIS; bedrock-runtime 'requests must name the geographic cross-Region inference ID us.openai.gpt-6.1-terra'."
      }
    ],
    "verified": "2026-11-17"
  }
}