Skip to content

models: gpt-5.6-sol is the default; decommission gpt-5.4-pro and older - #1

Open
espenale wants to merge 1 commit into
mainfrom
harness/decommission-old-gpt-models
Open

models: gpt-5.6-sol is the default; decommission gpt-5.4-pro and older#1
espenale wants to merge 1 commit into
mainfrom
harness/decommission-old-gpt-models

Conversation

@espenale

Copy link
Copy Markdown
Collaborator

What & why

InitialForce model policy (Espen, 2026-07-17): gpt-5.6-sol is the default oracle, and gpt-5.4-pro and everything older are decommissioned (removed from the OpenAI model list so they can't be selected).

@oysteinkrog — you MOD this fork (you added the 5.5/5.6-sol/terra/luna line), so flagging for your review before it goes live fleet-wide.

Changes

  • config.pyDEFAULT_MODEL fallback "auto""gpt-5.6-sol". Set DEFAULT_MODEL=auto (env/.env) to restore per-task auto-selection.
  • .env.exampleDEFAULT_MODEL=autogpt-5.6-sol (new deployments default correctly). Note: an existing local .env with DEFAULT_MODEL=auto still overrides — each machine updates its own .env (I've done mine).
  • conf/openai_models.json — removed 16 models: gpt-5.4-pro, gpt-5.4, gpt-5.3-codex, gpt-5.2-pro, gpt-5.2, gpt-5.1-codex, gpt-5.1-codex-mini, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-codex, gpt-4.1, o3, o3-mini, o3-pro, o4-mini. Kept the current generation: gpt-5.5, gpt-5.5-pro, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna. Gemini / xAI / other providers untouched.
  • .claude/CLAUDE.md — startup-guidance model examples updated off the decommissioned names.

Safety / verification

  • No PAL Python hardcodes a removed model name (grep-verified across src/), so nothing in the server breaks.
  • conf/openai_models.json is valid JSON; config.py imports and resolves; gpt-5.6-sol is present in the kept list.
  • The diff on the model file is deletions-only (kept entries byte-unchanged).

The one behavior change to weigh

The config.py default flips from auto (Claude picks per task) to fixed gpt-5.6-sol. If you'd rather keep auto-mode as the default and only prune the model list, drop the config.py/.env.example hunks. And adjust the decommission boundary if you want to keep, say, the codex line or o-series.

🤖 Generated with Claude Code

…ro and older

InitialForce policy (Espen, 2026-07-17): gpt-5.6-sol is the default oracle model, and gpt-5.4-pro and
everything older are decommissioned (removed from the OpenAI model list so they can't be selected).

- config.py: DEFAULT_MODEL fallback "auto" -> "gpt-5.6-sol" (set DEFAULT_MODEL=auto to restore per-task
  auto-selection). No PAL python hardcodes an old model name (grep-verified), so nothing breaks.
- conf/openai_models.json: removed gpt-5.4-pro, gpt-5.4, gpt-5.3-codex, gpt-5.2-pro, gpt-5.2,
  gpt-5.1-codex, gpt-5.1-codex-mini, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-codex, gpt-4.1, o3, o3-mini,
  o3-pro, o4-mini. KEPT the current generation: gpt-5.5, gpt-5.5-pro, gpt-5.6-sol, gpt-5.6-terra,
  gpt-5.6-luna. (Gemini/xAI/other providers untouched.)
- .claude/CLAUDE.md: startup-guidance model examples updated off the decommissioned names.

Scoped to the InitialForce fork. @oysteinkrog — you MOD this fork, so flagging for your review before it
goes live fleet-wide (the config.py default flip from auto->fixed is the main behavior change; adjust the
exact decommission cut if you want a different boundary).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@espenale
espenale requested a review from oysteinkrog July 17, 2026 11:47

@oysteinkrog oysteinkrog left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

None of this is strictly necessary..
When the server is installed on a machine there needs to be proper .env/configuration set no matter what. If we are going to customize default .env I think that might be a good idea but I would strongly recommend also disabling some of the tools which are not really needed and take up context.

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.

2 participants