Repository navigation
Codex Fast mode omits the GPT-6 Astra multiplier #1705
Description
What happened?
Codex reports charge GPT-6 Astra Fast-mode usage at the Standard rate. The live LiteLLM entry supplies token prices but no provider_specific_entry.fast, and ccusage has no Astra fallback multiplier, so it defaults to 1.0. Other supported Codex models already have explicit Fast multipliers.
Steps to reproduce
- Use Codex with
gpt-6-astraand Fast mode (service_tier = "priority"). - Compare
ccusage codex daily --speed standardandccusage codex daily --speed fastfor an Astra-only session. - The Astra cost stays the same.
Expected behavior
Apply the documented 2x API Fast multiplier for GPT-6 Astra: https://developers.openai.com/api/docs/models/gpt-6-astra
For example, 1M cached input tokens should have a $2.00 API-equivalent estimate in Fast mode instead of $1.00. I can submit a small fix using the existing multiplier table and regression tests.
Version
20.0.20; the missing multiplier is also present on main at 401a9a7.
Activity
pullfrog commentedon Sep 11, 2026
github-actions commentedon Sep 11, 2026
Automatic implementation was not started because the issue author GitHub email could not be resolved reliably for co-author attribution. A maintainer can implement the issue manually or provide a verifiable author email.
ryoppippi commentedon Sep 15, 2026
@pullfrog Please prepare a focused implementation for this bug. Reuse the existing Codex fast-mode multiplier path, add regression coverage for GPT-6 Astra with priority service tier, and keep standard-mode pricing unchanged. Verify the multiplier against the linked official OpenAI documentation; do not introduce a broader pricing fallback. Leave completed changes uncommitted in the working tree and return prepared metadata; the workflow will validate, commit with reporter co-authorship, push, and open the pull request.
pullfrog commentedon Sep 15, 2026
Muse Spark Contributor(free via Pullfrog for OSS) ｜ 𝕏ryoppippi commentedon Sep 15, 2026
Implementation is in #1726. One pricing detail from the original report needed correction: at 1M cached input tokens, GPT-6 Astra uses the documented long-context tier, so Standard is $2 and Fast is $4. The PR now covers that path and also preserves the 2x Fast multiplier when pricing falls back to models.dev.