Skip to content

Implementation Plan: T5-P4-A5-WP1 Replace three-way gpt-5.5 collision in CODEX_MODEL_ALIASES#4097

Merged
Trecek merged 3 commits into
developfrom
t5-p4-a5-wp1-replace-the-three-way-gpt-5-5-collision-in-code/4022
Jun 12, 2026
Merged

Implementation Plan: T5-P4-A5-WP1 Replace three-way gpt-5.5 collision in CODEX_MODEL_ALIASES#4097
Trecek merged 3 commits into
developfrom
t5-p4-a5-wp1-replace-the-three-way-gpt-5-5-collision-in-code/4022

Conversation

@Trecek

@Trecek Trecek commented Jun 12, 2026

Copy link
Copy Markdown
Collaborator

Summary

Replace the degenerate three-way gpt-5.5 collision in CODEX_MODEL_ALIASES with three distinct real OpenAI model IDs aligned with CODEX_EFFORT_MAPPING semantics: gpt-5.5 (opus/xhigh), gpt-5.4 (sonnet/high), gpt-5.4-mini (haiku/medium). Implement a model_class() utility function for cross-backend opus/sonnet/haiku tier resolution via suffix stripping, identity check on CLAUDE_MODEL_ALIASES keys, and reverse-map on CODEX_MODEL_ALIASES values. Thread the new symbol through the re-export chain and add architecture and unit tests.

Implementation Plan

Plan file: /home/talon/projects/autoskillit-runs/impl-20260612-150756-734470/.autoskillit/temp/make-plan/t5_p4_a5_wp1_replace_gpt55_collision_plan_2026-06-12_151500.md

Closes #4022

🤖 Generated with Claude Code via AutoSkillit

Token Usage Summary

Step Model count uncached output cache_read peak_ctx turns cache_write time
plan* opus[1m] 1 67 22.1k 1.7M 106.3k 51 107.1k 11m 37s
verify* sonnet 1 78 16.1k 407.5k 61.2k 26 40.3k 6m 36s
implement* MiniMax-M3 1 1.1M 6.7k 0 0 50 0 2m 48s
audit_impl* sonnet 1 2.3k 6.3k 166.5k 41.7k 16 29.8k 3m 49s
prepare_pr* MiniMax-M3 1 232.8k 2.8k 0 0 15 0 1m 8s
compose_pr* MiniMax-M3 1 176.4k 1.3k 0 0 12 0 36s
review_pr* sonnet 1 134 34.3k 935.5k 86.9k 48 66.9k 7m 19s
resolve_review* sonnet 1 235 12.5k 1.8M 83.7k 68 144.0k 7m 17s
Total 1.5M 102.1k 5.0M 106.3k 388.1k 41m 13s

* Step used a non-Anthropic provider; caching behavior may differ.

Token Efficiency

Step LoC Changed cache_read/LoC cache_write/LoC output/LoC
plan 0
verify 0
implement 59 0.0 0.0 113.2
audit_impl 0
prepare_pr 0
compose_pr 0
review_pr 0
resolve_review 11 166826.0 13094.4 1135.0
Total 70 71611.8 5544.7 1458.1

Model Usage Breakdown

Model steps uncached output cache_read cache_write time
opus[1m] 1 67 22.1k 1.7M 107.1k 11m 37s
sonnet 4 2.8k 69.2k 3.3M 281.0k 25m 3s
MiniMax-M3 3 1.5M 10.8k 0 0 4m 32s

Trecek and others added 3 commits June 12, 2026 15:28
Replace the degenerate three-way gpt-5.5 collision in CODEX_MODEL_ALIASES with
three distinct real OpenAI model IDs aligned with CODEX_EFFORT_MAPPING semantics:
- gpt-5.5 (opus/xhigh)
- gpt-5.4 (sonnet/high)
- gpt-5.4-mini (haiku/medium)

Add a model_class() utility function for cross-backend opus/sonnet/haiku tier
resolution via suffix stripping, identity check on CLAUDE_MODEL_ALIASES keys,
and reverse-map on CODEX_MODEL_ALIASES values. Thread the new symbol through
the re-export chain and add architecture and unit tests.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Guard against silent silent-drop in _CODEX_MODEL_REVERSE dict comprehension
when duplicate values in CODEX_MODEL_ALIASES would cause one alias to become
unreachable via model_class().

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…nst TypeError

Backends that cannot be default-constructed would raise TypeError rather than
a meaningful assertion failure. Skip such backends with an explanatory message
so the test degrades gracefully instead of masking the real invariant.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@Trecek Trecek added this pull request to the merge queue Jun 12, 2026
Merged via the queue into develop with commit 64227bd Jun 12, 2026
3 checks passed
@Trecek Trecek deleted the t5-p4-a5-wp1-replace-the-three-way-gpt-5-5-collision-in-code/4022 branch June 12, 2026 23:08
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