Skip to content

v2.18.0

Choose a tag to compare

View all tags
@ndycode ndycode released this 29 Sep 23:19
· 143 commits to main since this release
v2.18.0
328343b

https://www.npmjs.com/package/codex-multi-auth/v/2.18.0

npm i -g codex-multi-auth@2.18.0

Model-catalog release tracking upstream openai/codex commit b1e72963 (2026-09-29, #49318), which made GPT-6.1 Sol the default catalog model — priority 1, above Astra, "near-Astra performance at a lower cost". The change also stages the gpt-5.5 family retirement two weeks ahead of its 2026-10-14 OAuth shutdown and retires the leaked gpt-6-astra-aeon slug. (#725)

Added

  • gpt-6.1-sol support end-to-end and as the new DEFAULT_MODEL: model-map profile (low through ultra, default low), effort aliases plus bare gpt-6.1, a dedicated 6.1 branch in the GPT-6 resolver, capability metadata, config template picker entries, and a mirrored implementation in the standalone codex-multi-auth-codex wrapper. (#725)
  • Published gpt-6.1-sol rate card in the usage ledger: $2 / $10 per 1M in/out, $0.10 cached input, the >272K long-context tier ($4 / $15 / $0.20), and the 2x Fast tier. (#725)
  • New test/gpt61-sol-models.test.ts suite covering aliases, reasoning ladders, pricing, context windows, and the fallback walk. (#725)

Changed

  • gpt-5.5 and gpt-5.5-pro — including dated snapshot ids — resolve to gpt-6-sol and gpt-6-astra, upstream's named migration targets ahead of the 2026-10-14 OAuth retirement. Retired ids take the replacement's reasoning ladder (none/minimal coerce up to low) and price through its rate card. (#725)
  • gpt-6-astra-aeon resolves to gpt-6-astra; the leaked long-horizon slug never appeared in the durable bundled catalog. (#725)
  • Unsupported-model fallback chain re-floored off gpt-5.5: gpt-6.1-sol → gpt-6-sol → gpt-5.6-sol → gpt-6-luna (the broadest-availability catalog model), with the pro tier stepping gpt-6-astra → gpt-6-sol first and retired ids entering the same staircase via RETIRED_MODEL_REPLACEMENTS. The one-hop-at-a-time walk and per-request attempt budget are unchanged. (#725)
  • gpt-5 and unknown GPT-5-flavored ids land on gpt-5.6-sol; gpt-5.4 routes to gpt-6-sol, gpt-5.4-mini/nano to gpt-6-luna, and gpt-5.4-pro to gpt-6-astra. (#725)
  • Context-window estimates emptied deliberately: models with uncertain Codex-vs-API limits (gpt-6.1-sol, gpt-6-*, gpt-5.6-*, Daybreak) return null without a user override instead of evaluating sessions against a fabricated window. (#725)
  • Wrapper forward bound raised from 4 to 5 spawns so the deepest supported-model staircase completes on the last allowed forward — the same bound the plugin host's shared outbound-attempt budget uses. (#725)

Fixed

  • Usage pricing resolves bare aliases (gpt-6.1, gpt-6, gpt-5.6, astra) and effort-suffixed ids (gpt-6.1-sol-max) through the exact alias map, so raw ledger rows price at the canonical card instead of reading unpriced and failing a maxCostUsd budget closed. The fuzzy resolver is deliberately not used — unknown ids still price as unknown, never at the default's rate. (#725)
Assets 2
Loading