Skip to content

Move the CI agent model to gpt-5.6-terra - #31

Merged
dimavrem22 merged 1 commit into
mainfrom
chore/gpt-5.6-terra
Jul 28, 2026
Merged

dimavrem22 merged 1 commit into
mainfrom
chore/gpt-5.6-terra

Conversation

@dimavrem22

Copy link
Copy Markdown
Contributor

Summary

Moves every live-suite agent model pin to gpt-5.6-terra.

The workflows pinned two different models — gpt-5.4 on the a2a and channels suites, gpt-5.4-mini on voice and external-events. gpt-5.6-terra is the current mini tier and a generation newer than both, so that split stops buying anything and all four suites converge on one model.

value
Model id gpt-5.6-terra
Tier mini ("roughly corresponds to the mini model tier used in earlier GPT-5 families")
Context 1,050,000 (max input 922,000)
Max output 128,000
Price $2.50 in / $15 out / $0.25 cached, per 1M

Caveat worth recording

Realtime is not a supported endpoint for terra (nor Assistants, fine-tuning, embeddings, or the audio/image generation surfaces; Chat Completions and Batch are supported).

This does not affect the voice suite — it drives Realtime through gpt-realtime-2, a separate model from the agent's chat model — but it rules terra out for anything pointed at a realtime path.

Scope

CI workflow pins only. No plugin runtime, no gateway, no API change, no version bump.

Deliberately not included: the Delphi Hermes gateway default in servers (infra/variables.tf), which moved to gpt-5.6-sol earlier today for price parity with 5.5. That is a production agent on the full tier; dropping it to mini is a product decision rather than a CI cleanup.

Source: GPT-5.6 Terra model card

The live suites pinned gpt-5.4 and gpt-5.4-mini. gpt-5.6-terra is the
current mini-tier model -- a generation newer than both, with a 1.05M
context window against $2.50/$15 per 1M tokens -- so the split between
the two old pins no longer buys anything and every suite moves to the
same model.

Realtime is not a supported endpoint for terra. That does not affect the
voice suite, which drives Realtime through gpt-realtime-2 rather than the
agent's chat model, but it rules terra out for any realtime path.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@dimavrem22
dimavrem22 marked this pull request as ready for review July 28, 2026 21:44
@dimavrem22
dimavrem22 merged commit 11cd02c into main Jul 28, 2026
20 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant