[OPIK-7392] [BE] fix: split cipx cache-creation tokens by TTL (5m vs 1h) - #7565
Merged
BorisTkachenko merged 3 commits intoJul 22, 2026
Merged
Conversation
Cache writes are billed by TTL (5-minute = 1.25x input, 1-hour = 2x input) and Claude Code writes its cache with a 1-hour TTL, but cipx_spends collapsed the split into a single u_cache_creation lump, so AI-spend priced everything at the 5m rate (~18% cost understatement). - migration 000109: add u_cache_creation_5m / u_cache_creation_1h to cipx_spends - CipxSpendDAO: parse and persist the split from usage.cache_creation - CipxSpendBlockDAO: a span is a single LLM call with one TTL, so each write block inherits the span's TTL into its tier (cache_creation_5m / _1h) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
java
Contributor
⏱️ pre-commit per-hook timing
⏭️ 40 skipped (no matching files changed)
|
BorisTkachenko
marked this pull request as ready for review
July 22, 2026 11:16
![baz-reviewer[bot]](https://avatars.githubusercontent.com/in/933528?s=60&v=4){kind=small}
LifeXplorer
reviewed
Jul 22, 2026
|
|
||
| -- Anthropic bills cache writes by TTL: 5-minute at 1.25x input, 1-hour at 2x input. Claude Code | ||
| -- writes its stable prefix with a 1-hour TTL, so pricing every cache-creation token at the 5m rate | ||
| -- understates cost (~18% in practice). The proxy already emits the split on |
Contributor
There was a problem hiding this comment.
Doubt that these kind of comments are valid for a DAO object in Opik repo :)
"understates cost (~18% in practice)"
Moreover IMO in OSS repo we should try to be as mean with cost explaining comments as possible
Contributor
Author
There was a problem hiding this comment.
Fixed in a44dc78 — trimmed the migration comment to the technical schema change (what the split columns are and that the proxy emits them), dropping the pricing/cost rationale. That business context lives in the cost-api repo where the pricing actually happens. Good call keeping the OSS side lean.
🤖 Reply posted via /address-github-pr-comments
Contributor
There was a problem hiding this comment.
Commit a44dc78 addressed this comment by removing the detailed cost/pricing explanation from the DAO migration comment while retaining only the relevant technical context.
Address PR #7565 review: keep the OSS migration comment to the technical schema change (what the columns are and their source), dropping the pricing/cost explanation, which lives in the cost-api repo. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Details
Persists the 5-minute vs 1-hour cache-creation split so AI-spend can price 1h cache writes at 2× input. Previously
cipx_spendscollapsed the split into a singleu_cache_creationlump, so cost was computed at the 5m rate (1.25× input) for all cache-creation tokens — a ~18% understatement, since Claude Code writes its cache with a 1-hour TTL (2× input).000109: addu_cache_creation_5m/u_cache_creation_1htocipx_spendsCipxSpendDAO: parse and persist the split fromusage.cache_creationCipxSpendBlockDAO: a cipx span is a single LLM call with one cache TTL, so each write block inherits the span's TTL into itstier(cache_creation_5m/cache_creation_1h); legacy rows keepcache_creationand are read as 1h downstreamChange checklist
Issues
AI-WATERMARK
AI-WATERMARK: yes
/usageTesting
CostIntelligenceIngestionTestextended with 5m/1h split assertions and a parameterized write-block TTL-inheritance test.cipx_spends.u_cache_creation_1handcipx_spend_blocks.tier='cache_creation_1h'populate correctly, and the AI-spend UI total matches/usageto the cent.Paired cost-api change that applies the pricing: comet-ml/ai-cost-backend#36
Important
Deploy order: this migration (000109) must be deployed before the cost-api change, which reads the new columns.
Documentation
N/A