1{2 "schema_version": 2,3 "provider": "openai",4 "verified_at": "2026-08-21",5 "verified_region": "us-east-2",6 "source": "AWS Bedrock userguide OpenAI model cards + models-api-compatibility (retrieved 2026-08-21); internal migration reference, OpenAI path (feature-level probes, 2026-06-20)",7 "disclaimer": "This is the OpenAI-to-Bedrock MIGRATION target catalog, not a list of OpenAI models: it holds the verified landing candidates for the available paths. Verified 2026-08-21 — the GPT-5.6 family changed materially between 2026-07-21 and 2026-08-21: Sol/Terra/Luna now ALSO run on bedrock-runtime via CRIS inference profiles (Geo us./in., Global global.), with Converse/Chat Completions/Invoke listed as supported there (Guardrails Converse-only, prompt caching Responses-only on runtime), and Global CRIS pricing published at OpenAI's standard list price (cost parity; In-Region/Geo remain 1.10x). So runtime_converse now carries BOTH same-vendor GPT-5.6 candidates (same-model governance path) AND the Claude tier mapping (GPT-5.6 Sol -> Claude Opus 4.8, Terra/5.5/5.4 -> Claude Sonnet 5, Luna -> Claude Haiku 4.5) as the cross-family option. GPT-5.5 and GPT-5.4 remain mantle-only and in-region only — no CRIS, no runtime path. The gpt-oss open-weight models still support Converse on bedrock-runtime but sit a capability class below the frontier tier and are not listed. Only models with dated evidence are listed; entries name their evidence. Numeric limits stay 'unknown' unless sourced. Re-probe every selected model and API path in the target account and region before implementation."
18 "Preserves the OpenAI SDK and minimizes client rewrite",
19 "GPT-5.x on Bedrock Mantle is served through the Responses API"
20 ],
21 "constraints": [
22 "Mantle itself has no cross-region form for any GPT model (in-region only); for GPT-5.6 the bedrock-runtime CRIS path covers additional regions instead (see runtime_converse)",
23 "GPT-5.x on Mantle rejects Chat Completions; a Chat Completions source must reshape to Responses (reference-verified on gpt-5.4; the 5.6 cards list Chat Completions as a supported API but its mantle-side availability is unprobed — verify in the target account)",
24 "Availability must be probed in the target account and region"
53 "Supports Bedrock Guardrails (Converse API only), invocation logging, and CloudWatch",
54 "For GPT-5.6 this is a SAME-MODEL governance path via CRIS (us./in./global. ids; base URL bedrock-runtime.{region}.amazonaws.com/openai/v1 for the OpenAI-compatible APIs); the model cards recommend runtime for new applications, and Global CRIS is priced at OpenAI's standard rate (cost parity)"
55 ],
56 "constraints": [
57 "GPT-5.5 / GPT-5.4 have NO runtime path — a Converse target for them means a Bedrock-native model and a boto3 rewrite of the OpenAI integration",
58 "GPT-5.6 on runtime is CRIS-only (no in-region form); Geo CRIS stays within the geography, Global CRIS routes anywhere — pick by data-residency posture. Runtime does not support server-side tool use, structured outputs, or application inference profiles for these models; prompt caching is Responses-only",
59 "Bedrock-native models generally require a Global or geography-scoped CRIS inference profile",
60 "Availability must be probed in the target account and region"
61 ]
62 }
63 },
64 "models": {
65 "openai_gpt_5_6_sol": {
66 "display_name": "OpenAI GPT-5.6 Sol (on Bedrock)",
67 "family": "openai_gpt_5",
68 "generation": "reasoning",
69 "version": "5.6",
70 "class": "reasoning",
71 "context_window": 1000000,
72 "output_token_ceiling": "unknown",
73 "region_availability": {
74 "commercial": ["us-east-1", "us-east-2"],
75 "govcloud": [],
76 "evidence": "Commercial set per this catalog's pricing cache (US East N. Virginia / Ohio). The 2026-08 GovCloud announcement names Terra and Luna only \u2014 Sol's GovCloud availability is not stated; treat as unknown, not absent."
77 },
78 "capabilities": [
79 "reasoning",
80 "tool_or_function_calling",
81 "image_input_vision"
82 ],
83 "capability_source": "AWS model card 2026-08-21 (launch 2026-07-13, Active): frontier reasoning/agentic, image+text input checked; tool calling via the Responses API's built-in tool use (models-api-compatibility). Structured output/streaming behavior not feature-probed on Mantle — verify before relying. Re-verified 2026-08-21: context window is 1M per the current card (272K was the 2026-07-21 value); long-context (>272K) is a separate price tier at 2.0x input / 1.5x output per inference option.",
84 "paths": {
85 "mantle_openai_responses": {
86 "available": true,
87 "model_id": "openai.gpt-5.6-sol",
88 "requires_cris": false,
89 "evidence": "AWS model card (2026-08-21): bedrock-mantle at /openai/v1, in-region only; Responses supported; In-Region pricing at 1.10x OpenAI standard"
90 },
91 "mantle_openai_chat": {
92 "available": false,
93 "model_id": "openai.gpt-5.6-sol",
94 "requires_cris": false,
95 "evidence": "AWS model card: Chat Completions unchecked for this model"
96 },
97 "runtime_converse": {
98 "available": true,
99 "model_id": "openai.gpt-5.6-sol",
100 "requires_cris": true,
101 "evidence": "AWS model card (2026-08-21): bedrock-runtime row with Geo (us.) and Global CRIS ids; APIs listed: Responses/Chat Completions/Invoke/Converse; Guardrails Converse-only, prompt caching Responses-only; Global CRIS priced at OpenAI standard (parity), In-Region/Geo at 1.10x"
102 }
103 }
104 },
105 "openai_gpt_5_5": {
106 "display_name": "OpenAI GPT-5.5 (on Bedrock)",
107 "family": "openai_gpt_5",
108 "generation": "reasoning",
109 "version": "5.5",
110 "class": "reasoning",
111 "context_window": 272000,
112 "output_token_ceiling": "unknown",
113 "capabilities": [
114 "reasoning",
115 "tool_or_function_calling"
116 ],
117 "capability_source": "AWS model card 2026-07-21 (launch 2026-06-01, Active): advanced coding/agentic tasks; tool calling via the Responses API's built-in tool use. Feature-level Mantle behavior not probed — verify before relying. Re-verified 2026-08-21: still mantle-only, in-region only; no CRIS, no runtime path, no long-context tier.",
118 "paths": {
119 "mantle_openai_responses": {
120 "available": true,
121 "model_id": "openai.gpt-5.5",
122 "requires_cris": false,
123 "evidence": "AWS model card model-card-openai-gpt-55 (2026-07-21): bedrock-mantle Programmatic Access with Responses sample code; Geo/Global inference not supported"
124 }
125 }
126 },
127 "openai_gpt_5_6_terra": {
128 "display_name": "OpenAI GPT-5.6 Terra (on Bedrock)",
138 "evidence": "AWS What's New 2026-08 (openai-gpt-terra-luna-govcloud): \"GPT-5.6 Terra and Luna are now generally available on Amazon Bedrock in AWS GovCloud (US-West) and AWS GovCloud (US-East)\"; commercial set per this catalog's pricing cache \u2014 verify against the pricing page before relying on it."
139 },
140 "capabilities": [
141 "reasoning",
142 "tool_or_function_calling"
143 ],
144 "capability_source": "AWS model card 2026-08-21: balanced performance at half GPT-5.5 cost; tool calling via the Responses API's built-in tool use. Re-verified 2026-08-21: context window is 1M per the current card (the 2026-07-21 card did not state it — cataloged as unknown then); long-context (>272K) is a separate price tier at 2.0x input / 1.5x output per inference option.",
145 "paths": {
146 "mantle_openai_responses": {
147 "available": true,
148 "model_id": "openai.gpt-5.6-terra",
149 "requires_cris": false,
150 "evidence": "AWS model card (2026-08-21): bedrock-mantle at /openai/v1, in-region only; Responses supported; In-Region pricing at 1.10x OpenAI standard"
151 },
152 "runtime_converse": {
153 "available": true,
154 "model_id": "openai.gpt-5.6-terra",
155 "requires_cris": true,
156 "evidence": "AWS model card (2026-08-21): bedrock-runtime row with Geo (us./in.) and Global CRIS ids; APIs listed: Responses/Chat Completions/Invoke/Converse; Guardrails Converse-only, prompt caching Responses-only; Global CRIS priced at OpenAI standard (parity), In-Region/Geo at 1.10x"
157 }
158 }
159 },
160 "openai_gpt_5_6_luna": {
161 "display_name": "OpenAI GPT-5.6 Luna (on Bedrock)",
171 "evidence": "AWS What's New 2026-08 (openai-gpt-terra-luna-govcloud): \"GPT-5.6 Terra and Luna are now generally available on Amazon Bedrock in AWS GovCloud (US-West) and AWS GovCloud (US-East)\"; commercial set per this catalog's pricing cache \u2014 verify against the pricing page before relying on it."
172 },
173 "capabilities": [
174 "reasoning",
175 "tool_or_function_calling"
176 ],
177 "capability_source": "AWS model card 2026-08-21: fastest/lowest-cost OpenAI tier for high-volume tasks; tool calling via the Responses API's built-in tool use. Re-verified 2026-08-21: context window is 1M per the current card (the 2026-07-21 card did not state it — cataloged as unknown then); long-context (>272K) is a separate price tier at 2.0x input / 1.5x output per inference option.",
178 "paths": {
179 "mantle_openai_responses": {
180 "available": true,
181 "model_id": "openai.gpt-5.6-luna",
182 "requires_cris": false,
183 "evidence": "AWS model card (2026-08-21): bedrock-mantle at /openai/v1, in-region only; Responses supported; In-Region pricing at 1.10x OpenAI standard"
184 },
185 "runtime_converse": {
186 "available": true,
187 "model_id": "openai.gpt-5.6-luna",
188 "requires_cris": true,
189 "evidence": "AWS model card (2026-08-21): bedrock-runtime row with Geo (us./in.) and Global CRIS ids; APIs listed: Responses/Chat Completions/Invoke/Converse; Guardrails Converse-only, prompt caching Responses-only; Global CRIS priced at OpenAI standard (parity), In-Region/Geo at 1.10x"
190 }
191 }
192 },
193 "openai_gpt_5_4": {
194 "display_name": "OpenAI GPT-5.4 (on Bedrock)",
195 "family": "openai_gpt_5",
196 "generation": "reasoning",
197 "version": "5.4",
198 "class": "reasoning",
199 "context_window": "unknown",
200 "output_token_ceiling": "unknown",
201 "capabilities": [
202 "tool_or_function_calling",
203 "structured_output_json",
204 "reasoning",
205 "streaming"
206 ],
207 "capability_source": "internal reference smoke tests and its reasoning-API migration notebook (Responses path, 2026-06 probes) — the only entry with feature-level Mantle behavior probes (tools, structured output, sampling); limits not stated Re-verified 2026-08-21: still mantle-only, in-region only; no CRIS, no runtime path, no long-context tier.",
219 "evidence": "internal reference verified gotcha: GPT-5.x on Mantle is Responses-only; Chat Completions rejected"
220 }
221 }
222 },
223 "anthropic_claude_sonnet_5": {
224 "display_name": "Claude Sonnet 5 (Converse tier match for GPT-5.6 Terra / 5.5 / 5.4)",
225 "family": "anthropic_claude",
226 "generation": "bedrock_native",
227 "version": "5",
228 "class": "runtime_native",
229 "context_window": 1000000,
230 "output_token_ceiling": 128000,
231 "capabilities": [
232 "tool_or_function_calling",
233 "reasoning",
234 "image_input_vision"
235 ],
236 "capability_source": "AWS model card model-card-anthropic-claude-sonnet-5 (2026-07-21): Converse checked, 1M context / 128K output, adaptive thinking always on, image input checked (tool_use/thinking/vision mapped to this provider's feature vocabulary). Balanced Claude tier — the Converse target for GPT-5.6 Terra, GPT-5.5, GPT-5.4, and unmatched/legacy sources.",
237 "paths": {
238 "runtime_converse": {
239 "available": true,
240 "model_id": "anthropic.claude-sonnet-5",
241 "requires_cris": true,
242 "evidence": "AWS model card (2026-07-21): Converse on bedrock-runtime checked; Geo (us/eu/au) and Global CRIS supported"
243 }
244 }
245 },
246 "anthropic_claude_opus_4_8": {
247 "display_name": "Claude Opus 4.8 (Converse tier match for GPT-5.6 Sol)",
248 "family": "anthropic_claude",
249 "generation": "bedrock_native",
250 "version": "4.8",
251 "class": "runtime_native",
252 "context_window": 1000000,
253 "output_token_ceiling": 128000,
254 "capabilities": [
255 "tool_or_function_calling",
256 "reasoning",
257 "image_input_vision"
258 ],
259 "capability_source": "anthropic-bedrock-2026-07-21 catalog (runtime_converse available; tool_use/extended_thinking/vision mapped to this provider's feature vocabulary). Frontier Claude tier — the Converse target for GPT-5.6 Sol.",
270 "display_name": "Claude Haiku 4.5 (Converse tier match for GPT-5.6 Luna)",
271 "family": "anthropic_claude",
272 "generation": "bedrock_native",
273 "version": "4.5",
274 "class": "runtime_native",
275 "context_window": 200000,
276 "output_token_ceiling": 64000,
277 "capabilities": [
278 "tool_or_function_calling",
279 "image_input_vision"
280 ],
281 "capability_source": "anthropic-bedrock-2026-07-21 catalog (runtime_converse available; tool_use/vision mapped to this provider's feature vocabulary; no extended-thinking evidence, so reasoning-requiring workloads fall back to a higher Claude tier). Fast/low-cost Claude tier — the Converse target for GPT-5.6 Luna.",