Skip to content

fix(pricing): map opus-4-7 ids to claude-opus-4-7 instead of legacy claude-opus-4 - #580

Merged
junhoyeo merged 1 commit into
junhoyeo:mainfrom
xczllgit:fix/opus-pricing-normalization
May 24, 2026
Merged

fix(pricing): map opus-4-7 ids to claude-opus-4-7 instead of legacy claude-opus-4#580
junhoyeo merged 1 commit into
junhoyeo:mainfrom
xczllgit:fix/opus-pricing-normalization

Conversation

@xczllgit

@xczllgit xczllgit commented May 21, 2026 •
edited by cubic-dev-ai Bot
Loading

Copy link
Copy Markdown
Contributor

Summary

  • normalize_model_name only had explicit branches for opus 4.5 and 4.6; anything else with "opus" + "4" fell through to a catch-all that returned claude-opus-4.
  • For ids like aws.claude-opus-4-7 this caused exact_match_openrouter to reach anthropic/claude-opus-4 via the model_part index — the legacy opus 4 entry priced at $15/$75 per M instead of opus 4.7's $5/$25.
  • Add an explicit 4.7 branch in front of 4.6 so opus-4-7 ids resolve to claude-opus-4-7. Catch-all is left in place to preserve the existing behavior tests rely on (e.g. test_normalize_opus_4_60_does_not_map_to_4_6).

Why this matters

A real session with 8.4M input / 873K output / 41.3M cache-read / 12.1M cache-write tokens on aws.claude-opus-4-7 was being billed at ~$480 instead of the correct ~$160 — a roughly 3x overcharge — because the resolver picked the legacy opus 4 price table.

How I traced it

The lookup chain for aws.claude-opus-4-7:

  1. aws. is not in PROVIDER_PREFIXES, so strip_known_provider_prefix skips it.
  2. No exact LiteLLM/OpenRouter match.
  3. normalize_version_separator produces aws.claude-opus-4.7 — still no match.
  4. normalize_model_name falls through to claude-opus-4 (the bug).
  5. exact_match_openrouter("claude-opus-4") reaches anthropic/claude-opus-4 via openrouter_model_part, returning the legacy $15/$75 entry.

After the fix, step 4 short-circuits to claude-opus-4-7, which exists in upstream LiteLLM data and prices correctly.

Relationship to #578

Independent. #578 normalizes anthropic/claude-{major}-{minor}-{family} (family-suffix order, e.g. anthropic/claude-4-6-sonnet) and adds 4.5/4.6 aliases. It does not touch normalize_model_name, has no 4.7 entries, and aws.-prefixed ids never reach its new code path. The two fixes don't overlap and won't conflict at merge.

Test plan

  • cargo test -p tokscale-core --lib pricing::lookup — 129 passed, 0 failed (3 new tests included)
  • cargo test -p tokscale-core — 674 + 10 + 3 + 0 passed across all suites, 0 failed
  • New regression tests:
    • test_normalize_opus_4_7_prefers_4_7_over_4 — short-form opus-4-7
    • test_normalize_opus_4_7_dot_prefers_4_7_over_4 — dot-form opus-4.7
    • test_aws_opus_4_7_does_not_degrade_to_opus_4 — the actual reported case, asserts both the resolved key (claude-opus-4-7) and that computed cost lands in the 4.7 band ($160) rather than legacy opus-4 band ($480)

🤖 Generated with Claude Code

Summary by cubic

Fix pricing normalization for Opus 4.7 so IDs resolve to claude-opus-4-7 instead of the legacy claude-opus-4. This corrects cost calculations for aws.claude-opus-4-7 and prevents ~3x overcharges.

  • Bug Fixes
    • Add explicit 4.7 branch in normalize_model_name (before 4.6) to route opus-4-7 and opus-4.7 to claude-opus-4-7.
    • Stop aws.claude-opus-4-7 from falling back to anthropic/claude-opus-4; add regression tests to verify the resolved key and 4.7 pricing.

Written for commit 6f05b84. Summary will update on new commits. Review in cubic

xiaochaozheng @claude
fix(pricing): resolve opus-4-7 ids to claude-opus-4-7 instead of clau…
6f05b84 
…de-opus-4

`normalize_model_name` only had explicit branches for opus 4.5 and 4.6
and fell through to a catch-all `else if contains_delimited_fragment("4")`
that returned `claude-opus-4`. For ids like `aws.claude-opus-4-7` the
catch-all swallowed 4.7, after which `exact_match_openrouter` reached
`anthropic/claude-opus-4` via the `model_part` index — i.e. the legacy
opus 4 entry priced at $15/$75 per M instead of the 4.7 entry at $5/$25.
Real-world impact: a session with 8.4M input / 873K output / 41.3M
cache-read / 12.1M cache-write tokens billed at ~$480 instead of ~$160,
roughly a 3x overcharge.

Add an explicit 4.7 branch in front of 4.6 so the opus normalizer
short-circuits to `claude-opus-4-7` before reaching the catch-all.
Future minor versions (4.8+) will still need their own branches; the
catch-all is left in place to preserve the existing behavior tests
(`test_normalize_opus_4_60_does_not_map_to_4_6` etc.) rely on.

Add three regression tests:
- `test_normalize_opus_4_7_prefers_4_7_over_4` — short-form `opus-4-7`
- `test_normalize_opus_4_7_dot_prefers_4_7_over_4` — dot-form `opus-4.7`
- `test_aws_opus_4_7_does_not_degrade_to_opus_4` — the actual reported
  case, asserting the resolved key and the computed cost falls in the
  4.7 price band (~$160) rather than the legacy opus-4 band (~$480).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
@vercel

vercel Bot commented May 21, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment
Project Deployment Actions Updated (UTC)
tokscale Ignored Ignored Preview May 21, 2026 7:41am

Request Review

cubic-dev-ai[bot]
cubic-dev-ai Bot reviewed May 21, 2026
View reviewed changes

@cubic-dev-ai cubic-dev-ai Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No issues found across 1 file

Re-trigger cubic

@junhoyeo

junhoyeo commented May 24, 2026

Copy link
Copy Markdown
Owner

Thanks!

@junhoyeo
junhoyeo merged commit 4bfa586 into junhoyeo:main May 24, 2026
14 checks passed
@junhoyeo

junhoyeo commented May 28, 2026

Copy link
Copy Markdown
Owner

@xczllgit this has been merged to v3.0.0: https://github.com/junhoyeo/tokscale/releases/tag/v3.0.0 thanks for the contribution!

@crhan crhan mentioned this pull request May 29, 2026
Closed
@xczllgit xczllgit mentioned this pull request May 29, 2026
Merged
3 tasks
junhoyeo added a commit that referenced this pull request Jun 10, 2026
@xczllgit xiaochaozheng
@claude @junhoyeo
fix(pricing): parse modern Claude version from id instead of hardcodi…
aebe4ea 
…ng (#634)

* fix(pricing): parse modern Claude version from id instead of hardcoding

`normalize_model_name` matched each opus/sonnet/haiku minor version with a
hardcoded branch and a catch-all `claude-opus-4`. Every new release (4.5 →
4.6 → 4.7 → 4.8) needed a new branch; a missing one let ids like
`aws.claude-opus-4-8` fall through to the catch-all, which OpenRouter then
resolved to legacy `anthropic/claude-opus-4` ($15/$75 per M instead of
$5/$25) — a ~3x overcharge. This repeated for 4.7 (#580) and again for 4.8.

The modern Claude line (major >= 4) follows the regular
`claude-{family}-{major}[-{minor}]` scheme, so parse the version straight
from the id and build the canonical key dynamically. New minor releases
(4.9, 5.0, …) now resolve to their own pricing key with no code change —
pricing values still come entirely from the upstream datasets.

Boundary contract from the old matcher is preserved: `opus-4-60` →
`claude-opus-4` (two-digit minor degrades to major), `opus-14-6` → None
(two-digit major is not the modern line), undelimited `opus4`/`opus-4x` →
None. The irregular legacy 3.x naming keeps its explicit branches.

Tests:
- test_normalize_future_minor_versions_resolve_without_hardcoding — 4.9/5.0
  and cross-family (sonnet/haiku) parse without a hardcoded entry
- test_normalize_modern_claude_boundaries — locks the boundary contract
- 4.8 regression tests mirroring the 4.7 set (short/dot/aws forms + cost band)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* style: wrap haiku condition to satisfy cargo fmt

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* fix(pricing): enforce never-degrade across all Claude families and majors

Generalize the opus-only version handling in PricingLookup so every
Claude family (opus, sonnet, haiku, fable) gets the same guarantees:

- normalize_claude_opus_4_minor -> normalize_claude_family_minor:
  family/major/minor parsed from the id in both orders (sonnet-4.7,
  claude-4-6-sonnet), so reversed-order sonnet/haiku ids resolve to
  their canonical key instead of cross-family fuzzy fallbacks.
- normalize_claude_family_bare_major: bare modern majors (claude-opus-5,
  claude-sonnet-5, fable-5) normalize to a canonical key that resolves
  only via an exact dataset hit.
- contains_delimited_modern_major_minor: the 4.x major-minor None-guard
  in normalize_model_name now covers majors 4-9, so 4-60 / 5-0 / dated
  forms stay unresolved instead of degrading.
- resolves_unsafe_claude_version (was resolves_different_claude_opus_4_
  minor): vetoes cross-family resolutions (bedrock sonnet ids billed as
  opus), cross-version resolutions for any family (sonnet-4-7 ->
  sonnet-4.6, haiku-4-6 -> haiku-4.5 / 3.5-haiku), and any modern-Claude
  resolution for ids whose version could not be parsed. Bare major 4
  stays unpinned to preserve existing claude-opus-4 fuzzy behavior.

Whether a version is "known" remains dataset-driven, exactly as it was
for opus: the canonical key either exact-matches the pricing dataset or
the id resolves unpriced. No hardcoded minor lists were added.

Reworks PR #634 on top of main's mechanisms instead of merging its
parallel parser.

Constraint: unknown minors/majors must resolve unpriced, never to a cheaper or different price
Constraint: all of main's existing pricing tests must pass unchanged
Rejected: PR #634's original generic parser | degrades two-digit minors (opus-4-60 -> claude-opus-4) and drops reversed-order coverage
Rejected: extending the hardcoded 4.5/4.6/4.7 allowlist in has_unrecognized_claude_four_minor | recreates the new-minor-release maintenance trap PR #634 set out to remove
Rejected: pinning bare major 4 ids (claude-opus-4) to exact-only | changes long-standing dated/regional 4.x resolution covered by existing tests
Confidence: high
Scope-risk: moderate
Directive: requested_claude_version deliberately unpins bare major 4; do not "complete" it to all majors without auditing dated-id (claude-opus-4-20250514) and regional-key resolution
Not-tested: provider-hinted lookups (lookup_with_provider) against cross-family adversarial datasets; models.dev reseller keys that legitimately embed reversed-order ids (cortecs/claude-4-6-sonnet) are covered only via real-data probing

---------

Co-authored-by: xiaochaozheng <xiaochaozheng@meituan.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Junho Yeo <i@junho.io>
@xczllgit
xczllgit deleted the fix/opus-pricing-normalization branch June 11, 2026 01:50
pinion05 added a commit to pinion05/tokscale that referenced this pull request Jun 23, 2026
@pinion05 xiaochaozheng
@claude @junhoyeo
fix(pricing): parse modern Claude version from id instead of hardcodi…
0b3391e 
…ng (junhoyeo#634)

* fix(pricing): parse modern Claude version from id instead of hardcoding

`normalize_model_name` matched each opus/sonnet/haiku minor version with a
hardcoded branch and a catch-all `claude-opus-4`. Every new release (4.5 →
4.6 → 4.7 → 4.8) needed a new branch; a missing one let ids like
`aws.claude-opus-4-8` fall through to the catch-all, which OpenRouter then
resolved to legacy `anthropic/claude-opus-4` ($15/$75 per M instead of
$5/$25) — a ~3x overcharge. This repeated for 4.7 (junhoyeo#580) and again for 4.8.

The modern Claude line (major >= 4) follows the regular
`claude-{family}-{major}[-{minor}]` scheme, so parse the version straight
from the id and build the canonical key dynamically. New minor releases
(4.9, 5.0, …) now resolve to their own pricing key with no code change —
pricing values still come entirely from the upstream datasets.

Boundary contract from the old matcher is preserved: `opus-4-60` →
`claude-opus-4` (two-digit minor degrades to major), `opus-14-6` → None
(two-digit major is not the modern line), undelimited `opus4`/`opus-4x` →
None. The irregular legacy 3.x naming keeps its explicit branches.

Tests:
- test_normalize_future_minor_versions_resolve_without_hardcoding — 4.9/5.0
  and cross-family (sonnet/haiku) parse without a hardcoded entry
- test_normalize_modern_claude_boundaries — locks the boundary contract
- 4.8 regression tests mirroring the 4.7 set (short/dot/aws forms + cost band)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* style: wrap haiku condition to satisfy cargo fmt

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* fix(pricing): enforce never-degrade across all Claude families and majors

Generalize the opus-only version handling in PricingLookup so every
Claude family (opus, sonnet, haiku, fable) gets the same guarantees:

- normalize_claude_opus_4_minor -> normalize_claude_family_minor:
  family/major/minor parsed from the id in both orders (sonnet-4.7,
  claude-4-6-sonnet), so reversed-order sonnet/haiku ids resolve to
  their canonical key instead of cross-family fuzzy fallbacks.
- normalize_claude_family_bare_major: bare modern majors (claude-opus-5,
  claude-sonnet-5, fable-5) normalize to a canonical key that resolves
  only via an exact dataset hit.
- contains_delimited_modern_major_minor: the 4.x major-minor None-guard
  in normalize_model_name now covers majors 4-9, so 4-60 / 5-0 / dated
  forms stay unresolved instead of degrading.
- resolves_unsafe_claude_version (was resolves_different_claude_opus_4_
  minor): vetoes cross-family resolutions (bedrock sonnet ids billed as
  opus), cross-version resolutions for any family (sonnet-4-7 ->
  sonnet-4.6, haiku-4-6 -> haiku-4.5 / 3.5-haiku), and any modern-Claude
  resolution for ids whose version could not be parsed. Bare major 4
  stays unpinned to preserve existing claude-opus-4 fuzzy behavior.

Whether a version is "known" remains dataset-driven, exactly as it was
for opus: the canonical key either exact-matches the pricing dataset or
the id resolves unpriced. No hardcoded minor lists were added.

Reworks PR junhoyeo#634 on top of main's mechanisms instead of merging its
parallel parser.

Constraint: unknown minors/majors must resolve unpriced, never to a cheaper or different price
Constraint: all of main's existing pricing tests must pass unchanged
Rejected: PR junhoyeo#634's original generic parser | degrades two-digit minors (opus-4-60 -> claude-opus-4) and drops reversed-order coverage
Rejected: extending the hardcoded 4.5/4.6/4.7 allowlist in has_unrecognized_claude_four_minor | recreates the new-minor-release maintenance trap PR junhoyeo#634 set out to remove
Rejected: pinning bare major 4 ids (claude-opus-4) to exact-only | changes long-standing dated/regional 4.x resolution covered by existing tests
Confidence: high
Scope-risk: moderate
Directive: requested_claude_version deliberately unpins bare major 4; do not "complete" it to all majors without auditing dated-id (claude-opus-4-20250514) and regional-key resolution
Not-tested: provider-hinted lookups (lookup_with_provider) against cross-family adversarial datasets; models.dev reseller keys that legitimately embed reversed-order ids (cortecs/claude-4-6-sonnet) are covered only via real-data probing

---------

Co-authored-by: xiaochaozheng <xiaochaozheng@meituan.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Junho Yeo <i@junho.io>
t1000040 pushed a commit to tmobi-internal/tokscale that referenced this pull request Jun 30, 2026
@xczllgit xiaochaozheng
@claude @junhoyeo @leecoder
fix(pricing): parse modern Claude version from id instead of hardcodi…
d9f7aa2 
…ng (junhoyeo#634)

* fix(pricing): parse modern Claude version from id instead of hardcoding

`normalize_model_name` matched each opus/sonnet/haiku minor version with a
hardcoded branch and a catch-all `claude-opus-4`. Every new release (4.5 →
4.6 → 4.7 → 4.8) needed a new branch; a missing one let ids like
`aws.claude-opus-4-8` fall through to the catch-all, which OpenRouter then
resolved to legacy `anthropic/claude-opus-4` ($15/$75 per M instead of
$5/$25) — a ~3x overcharge. This repeated for 4.7 (junhoyeo#580) and again for 4.8.

The modern Claude line (major >= 4) follows the regular
`claude-{family}-{major}[-{minor}]` scheme, so parse the version straight
from the id and build the canonical key dynamically. New minor releases
(4.9, 5.0, …) now resolve to their own pricing key with no code change —
pricing values still come entirely from the upstream datasets.

Boundary contract from the old matcher is preserved: `opus-4-60` →
`claude-opus-4` (two-digit minor degrades to major), `opus-14-6` → None
(two-digit major is not the modern line), undelimited `opus4`/`opus-4x` →
None. The irregular legacy 3.x naming keeps its explicit branches.

Tests:
- test_normalize_future_minor_versions_resolve_without_hardcoding — 4.9/5.0
  and cross-family (sonnet/haiku) parse without a hardcoded entry
- test_normalize_modern_claude_boundaries — locks the boundary contract
- 4.8 regression tests mirroring the 4.7 set (short/dot/aws forms + cost band)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* style: wrap haiku condition to satisfy cargo fmt

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* fix(pricing): enforce never-degrade across all Claude families and majors

Generalize the opus-only version handling in PricingLookup so every
Claude family (opus, sonnet, haiku, fable) gets the same guarantees:

- normalize_claude_opus_4_minor -> normalize_claude_family_minor:
  family/major/minor parsed from the id in both orders (sonnet-4.7,
  claude-4-6-sonnet), so reversed-order sonnet/haiku ids resolve to
  their canonical key instead of cross-family fuzzy fallbacks.
- normalize_claude_family_bare_major: bare modern majors (claude-opus-5,
  claude-sonnet-5, fable-5) normalize to a canonical key that resolves
  only via an exact dataset hit.
- contains_delimited_modern_major_minor: the 4.x major-minor None-guard
  in normalize_model_name now covers majors 4-9, so 4-60 / 5-0 / dated
  forms stay unresolved instead of degrading.
- resolves_unsafe_claude_version (was resolves_different_claude_opus_4_
  minor): vetoes cross-family resolutions (bedrock sonnet ids billed as
  opus), cross-version resolutions for any family (sonnet-4-7 ->
  sonnet-4.6, haiku-4-6 -> haiku-4.5 / 3.5-haiku), and any modern-Claude
  resolution for ids whose version could not be parsed. Bare major 4
  stays unpinned to preserve existing claude-opus-4 fuzzy behavior.

Whether a version is "known" remains dataset-driven, exactly as it was
for opus: the canonical key either exact-matches the pricing dataset or
the id resolves unpriced. No hardcoded minor lists were added.

Reworks PR junhoyeo#634 on top of main's mechanisms instead of merging its
parallel parser.

Constraint: unknown minors/majors must resolve unpriced, never to a cheaper or different price
Constraint: all of main's existing pricing tests must pass unchanged
Rejected: PR junhoyeo#634's original generic parser | degrades two-digit minors (opus-4-60 -> claude-opus-4) and drops reversed-order coverage
Rejected: extending the hardcoded 4.5/4.6/4.7 allowlist in has_unrecognized_claude_four_minor | recreates the new-minor-release maintenance trap PR junhoyeo#634 set out to remove
Rejected: pinning bare major 4 ids (claude-opus-4) to exact-only | changes long-standing dated/regional 4.x resolution covered by existing tests
Confidence: high
Scope-risk: moderate
Directive: requested_claude_version deliberately unpins bare major 4; do not "complete" it to all majors without auditing dated-id (claude-opus-4-20250514) and regional-key resolution
Not-tested: provider-hinted lookups (lookup_with_provider) against cross-family adversarial datasets; models.dev reseller keys that legitimately embed reversed-order ids (cortecs/claude-4-6-sonnet) are covered only via real-data probing

---------

Co-authored-by: xiaochaozheng <xiaochaozheng@meituan.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Junho Yeo <i@junho.io>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@cubic-dev-ai cubic-dev-ai[bot] cubic-dev-ai[bot] left review comments
@junhoyeo junhoyeo junhoyeo approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@xczllgit @junhoyeo