Skip to content

GptTerra W1-01: Register GPT-6.1 Terra in the fork's Bedrock catalog - #39

Merged
samuelboland merged 1 commit into
mainfrom
dux-gptterra-w1-01-gpt-6-1-terra-registry
Nov 9, 2026
Merged

samuelboland merged 1 commit into
mainfrom
dux-gptterra-w1-01-gpt-6-1-terra-registry

Conversation

@staging-supernova-dx-appf-io

@staging-supernova-dx-appf-io staging-supernova-dx-appf-io Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Summary

Adds openai.gpt-6.1-terra to the Bedrock catalog so Supernova can offer it after a gem bump. Catalog and specs only: Providers::Bedrock::MANTLE_ONLY_MODEL_PATTERN, Protocols::MantleResponses::FRONTIER_GPT_PATTERN and Responses::Chat::GPT6_MODEL_PATTERN already match the id, so it routes to mantle /openai/v1/responses with the GPT-6 output-item handling.

  • models.json: a new entry after openai.gpt-6.1-sol.
    • Context window 1,050,000 and max output 128,000.
    • Efforts none low medium high xhigh max.
    • Modalities and capabilities as 6.1 Sol.
    • Standard-tier price per million tokens: 1.1 in, 5.5 out, 0.055 cache read, 1.375 cache write.
  • aliases.json: gpt-6.1-terra maps to openai.gpt-6.1-terra on bedrock.

Live checks (2026-11-09, account 436292035443, us-east-1 mantle)

  • Agreement: the Bedrock marketplace agreement is accepted, and a call returned HTTP 200.
  • Other routes: /v1/responses and /v1/chat/completions return 400, the same as 6.1 Sol.
  • Efforts: none through max return 200 and minimal returns 400. 6.1 Sol rejects none; Terra accepts it.
  • Context window: 921,007 input tokens are accepted and 923,500 are rejected (context_length_exceeded). That is 1,050,000 minus the 128,000 output reservation, as for Luna and both Sols.
  • Input types: image input works, and a PDF sent as input_file was read back correctly, so pdf stays.
  • Price: from the agreement offer's rate card. The *_standard dimensions are the US-only rate we are billed. global.* is blocked by an SCP. The long-context tier (2.2 / 8.25 / 0.11 / 2.75) is not stored, the same as for 6.1 Sol.

Tests

  • models_gpt_6_bedrock_spec.rb: Terra's efforts, cost and limits, plus the alias.
  • responses_gpt6_mantle_spec.rb: Terra is in gpt6_models with lowest_effort: 'none', sol: false, tool_loop: true, and seven cassettes were recorded live. The tool loop passes on Terra.
  • Routing lists: mantle_responses_spec.rb and bedrock_spec.rb include the new id.
  • Runs: each changed spec file ran three times with different seeds in replay mode, and RuboCop is clean. rake models validation reports 0 errors.
  • Cassettes: credentials are redacted (<AWS_ACCESS_KEY_ID>, <AWS_SESSION_TOKEN>).

Adds openai.gpt-6.1-terra to the model registry and aliases, routes it
through Mantle Responses, and records live cassettes for the GPT-6 suite.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Co-authored-by: Sam Boland <sam.boland@appfolio.com>
@samuelboland samuelboland changed the title feat: register openai.gpt-6.1-terra in Bedrock Mantle catalog GptTerra W1-01: Register GPT-6.1 Terra in the fork's Bedrock catalog Nov 9, 2026
@samuelboland
samuelboland marked this pull request as ready for review November 9, 2026 19:41
@samuelboland
samuelboland requested a review from a team as a code owner November 9, 2026 19:41

@samuelboland samuelboland left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed 558d9dd: the Terra entry matches the live measurements and the offer's US rate, the alias resolves, the routing lists include it, the seven recorded runs have credentials redacted, and CI is green.

@samuelboland
samuelboland merged commit 5b42e46 into main Nov 9, 2026
24 of 25 checks passed
@staging-supernova-dx-appf-io

Copy link
Copy Markdown
Author

🤖 Supernova Code Review — View trace

🔄 Re-run this review

— via automation "Code Review Opus 5.5 Medium" (run), installed by Sam Boland (sam.boland@appfolio.com)

@staging-supernova-dx-appf-io staging-supernova-dx-appf-io Bot left a comment

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I reviewed the PR and found nothing actionable.

  • models.json and aliases.json: the openai.gpt-6.1-terra entry has the same layout as openai.gpt-6.1-sol. Its pricing.text_tokens.standard values match metadata.cost (1.1 / 5.5 / 0.055 / 1.375). The efforts list includes none, which the live checks support. The context window and output limits match the other GPT-6 models. The alias is in alphabetical order.
  • Routing: MantleResponses::FRONTIER_GPT_PATTERN (/\Aopenai\.gpt-[56]/) matches the new id, so no code change is needed.
  • Specs: the new gpt6_models entry and the routing-list additions agree with the catalog. In the compaction example, (gpt6_models.keys - [model]).first picks openai.gpt-6-luna as Terra's switch target. Adding Terra at the end of the hash doesn't change the switch target for the existing models, so their cassettes still match.

🔄 Re-run this review

— via automation "Code Review Opus 5.5 Medium" (run), installed by Sam Boland (sam.boland@appfolio.com)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants