Epic: Support GPT-6.1 Terra (gpt-6.1-terra) #5558

Drafted by strandly-the-agent (AI) from the scheduled new-model watch on 2026-10-27; a human has not yet reviewed it.

OpenAI released GPT-6.1 Terra to the API on 2026-10-26 and AWS made it GA on Amazon Bedrock the same day. Neither id is in _CONTEXT_WINDOW_LIMITS / CONTEXT_WINDOW_LIMITS today, and the harness effort tables only match openai.gpt-5.6-* and openai.gpt-6-astra exactly.

Sources

• OpenAI model page: https://developers.openai.com/api/docs/models/gpt-6.1-terra
• OpenAI API changelog (Oct 26 entry, v1/responses + v1/chat/completions): https://developers.openai.com/api/docs/changelog
• Bedrock model card (launch date October 26, 2026): https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-6-1-terra.html
• AWS What's New: https://aws.amazon.com/about-aws/whats-new/2026/10/openai-gpt-6-1-terra-on-amazon-bedrock/

Model ids

• OpenAI API: gpt-6.1-terra (no dated snapshot published)
• Bedrock base id: openai.gpt-6.1-terra
• Bedrock inference profiles: us.openai.gpt-6.1-terra, global.openai.gpt-6.1-terra

Context window / max output

1,050,000 tokens context window / 128,000 max output tokens (same figures on the OpenAI model page and the Bedrock card). Knowledge cutoff June 1, 2026.

Capabilities

• Reasoning effort: none, low, medium (default), high, xhigh, max (OpenAI model page)
• Tools / function calling (Responses API), structured outputs, streaming, prompt caching, image input, computer use (OpenAI model page)
• Bedrock: bedrock-runtime Converse / Responses / Chat Completions / Invoke; bedrock-mantle under /openai/v1 (same path family as the other GPT-6.x ids, so _OPENAI_PATH_MODEL_PREFIXES already routes it); reasoning summaries; structured outputs (JSON Schema); implicit prompt caching on Converse / Invoke / Responses / Chat Completions, explicit caching on Responses only (Bedrock card)
• Pricing (OpenAI): $1.00 input / $0.05 cached / $5.00 output per 1M tokens

Release state

GA on both the OpenAI API and Amazon Bedrock since 2026-10-26.

Verification checklist

• Bedrock id heuristics for openai.gpt-6.1-terra: tool-result status inclusion (_MODELS_INCLUDE_STATUS / MODELS_INCLUDE_STATUS), prompt-cache strategy auto-detect, native count_tokens (_SKIP_COUNT_TOKENS_MODELS)
• Context-overflow and throttling error strings: confirm this model's overflow / throttle errors match BEDROCK_CONTEXT_WINDOW_OVERFLOW_MESSAGES (Converse) and the OpenAI / Mantle error mappers (_openai_errors.py)
• Re-check [BUG] Structured output fails with a 400 on Claude Opus 5.5 / Sonnet 5.5 / Fable 5.1 because the forced tool_choice fallback is rejected #5156 (structured output via forced tool_choice) against this model — passed on Bedrock Converse + conformance (End-to-end run: SDK integration tests against GPT-6.1 Terra (openai.gpt-6.1-terra, gpt-6.1-terra) #5560)
• Re-check Epic: Anthropic distillation protection (preserved thinking and reasoning_extraction refusals) #5160 (thinking / reasoning blocks replay) against this model — Converse: redacted reasoningContent is emitted and replays; Mantle Responses: emitted only at high effort, multi-turn replay still logs the known "not yet supported" warning (End-to-end run: SDK integration tests against GPT-6.1 Terra (openai.gpt-6.1-terra, gpt-6.1-terra) #5560)
• Optional (human call): default-model bumps (harness-py/src/strands_harness/models.py, harness-ts/src/models.ts, integ fixtures), docs examples

Sub-issues

• Context-window entry for gpt-6.1-terra / openai.gpt-6.1-terra (py + ts) #5559 — Context-window entry for gpt-6.1-terra / openai.gpt-6.1-terra (py + ts) — PR fix(models): add GPT-6.1 Terra context window (py + ts + cli) #5563
• End-to-end run: SDK integration tests against GPT-6.1 Terra (openai.gpt-6.1-terra, gpt-6.1-terra) #5560 — End-to-end run: SDK integration tests against GPT-6.1 Terra
• Harness: Bedrock effort levels do not apply to openai.gpt-6.1-terra (exact-match on openai.gpt-6-astra) #5561 — Harness: Bedrock effort levels do not apply to openai.gpt-6.1-terra (exact-match on openai.gpt-6-astra)

Not needed: Mantle path routing — openai.gpt-6. is already in _OPENAI_PATH_MODEL_PREFIXES / OPENAI_PATH_MODEL_PREFIXES, so openai.gpt-6.1-terra routes to /openai/v1 today.

(Tracked via task list: POST /issues/5558/sub_issues returned 404 for strandly-the-agent, so these are not formally attached — a maintainer can convert them.)