fix(pricing): refresh the rate card and add gemini 3.7/3.8 Flash - #155
Merged
Conversation
Re-verified every row against the vendor cards on 2026-09-03. Four rows were stale and three models were missing. Repriced: - claude-sonnet-5 $3/$15 -> $2/$10. The $2/$10 introductory rate is now the standard price and the 2026-09-01 increase was cancelled, so the deliberate hedge against it lapsing was overstating the nightly's own model by 50%. - gpt-5.6-sol $5/$30 -> $4/$20. Sol took its own cut after the 2026-07-30 Terra/Luna one. - z-ai/glm-5.2 and deepseek/deepseek-v4-pro, read from OpenRouter's live /api/v1/models feed. Both had drifted up; deepseek's cache read was low by ~24x. Added: - gemini-3.8-flash and gemini-3.7-flash. 3.8 is now the ANTIGRAVITY_MODEL in both variable stores, and an unpriced model books no cost at all, so an antigravity run reported cost_complete=false with a null bill. - claude-fable-5-1, which prices cache hits at 0.025x input rather than the 0.1x every other Claude model uses. The three Bedrock open-weight rows are carried forward untouched: AWS does not publish eu-north-1 figures for them, and the comment there warns against substituting the US column. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Drop the price-change narratives. What each row USED to cost is git history; the comments now carry only the traps a future edit can fall into. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
bai-uipath
marked this pull request as ready for review
September 3, 2026 21:10
bai-uipath
requested review from
CarlesUIPath,
akshaylive,
tmatup and
uipreliga
as code owners
September 3, 2026 21:10
|
Claude Code is working… I'll analyze this and get back to you. |
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.
Re-verified every row of the built-in rate card against the vendor cards, and added the models the antigravity bump now needs. Four rows were stale; three models were missing entirely.
Why now
ANTIGRAVITY_MODELwas just moved togemini-3.8-flashin both variable stores (theskillsrepo variable and the ADOcoder-eval-athena-configgroup). Antigravity computes cost locally from tokens against this table, so an unpriced id books no cost at all — the run still passes, but reportscost_complete: falseand a null bill. That prompted a full audit rather than just the one row.Repriced
claude-sonnet-5gpt-5.6-solz-ai/glm-5.2deepseek/deepseek-v4-proclaude-sonnet-5is the significant one. The table deliberately held $3/$15 against the $2/$10 introductory rate lapsing on 2026-08-31. Anthropic has since made $2/$10 the standard price and cancelled the scheduled increase, so that hedge was overstating every run of the nightly's own model by 50%.The two OpenRouter rows come from the live
/api/v1/modelsfeed. Both had drifted up, anddeepseek-v4-pro's cache-read rate was low by roughly 24x.Added
gemini-3.8-flashandgemini-3.7-flash— same list rate as 3.6.claude-fable-5-1— it prices cache hits at 0.025x input, not the 0.1x every other Claude model uses. Worth a row of its own so nobody copies Fable 5's figure onto it.Left alone, deliberately
The three Bedrock open-weight rows (
deepseek.v3.2,zai.glm-5,moonshotai.kimi-k2.5) are unchanged. AWS does not publish eu-north-1 figures for them, and the existing comment warns against substituting the US column. Carried forward and documented rather than guessed.Verification
Full Python suite, evalboard
tsc --noEmit, and all 608 evalboard tests pass, including thepricing.py↔pricing.tsparity guard. The 13 failures intest_judge_litellm.py/test_reports_stats_nonfinite.pyon my machine reproduce identically with these changes stashed — a local venv artifact, not this branch.The gemini rows were confirmed end to end: an antigravity run on
gemini-3.8-flashwent fromcost_complete: false/ null cost tocost_complete: true/ $0.0685 once the row was present.