fix(pricing): GPT-5.3 Codex resolves to $0.00 — filter github_copilot + inject overrides by junhoyeo · Pull Request #207 · junhoyeo/tokscale

junhoyeo · 2026-02-18T03:53:02Z

Summary

Filter github_copilot/ entries from LiteLLM data — subscription-based $0.00 pricing poisons per-token cost estimation (e.g. github_copilot/gpt-5.3-codex → $0.00/$0.00)
Inject hardcoded pricing overrides for gpt-5.3, gpt-5.3-codex, gpt-5.3-codex-spark at $1.75/$14.00 per 1M tokens ($0.175 cache read) — sourced from Cursor docs + llm-stats.com corroboration
Overrides use or_insert_with — they won't clobber future LiteLLM/OpenRouter entries once official API pricing lands (LiteLLM PR #20552 pending)
6 new unit tests covering filter, override injection, and lookup

Root Cause

tokscale pricing "gpt-5.3-codex" matched github_copilot/gpt-5.3-codex from LiteLLM data, which reports $0.00/$0.00 (subscription-based, meaningless for cost estimation). GPT-5.3 Codex is not yet on the OpenAI API, so no legitimate entry exists in either LiteLLM or OpenRouter.

Test Results

cargo test — 166 passed, 0 failed, 1 ignored (pre-existing sqlite test)

Closes #199, Closes #172, Closes #204, Closes #205, Closes #206

Summary by cubic

Fixes $0.00 pricing for GPT-5.3 Codex by filtering subscription-only GitHub Copilot entries and adding Cursor-priced overrides for the GPT-5.3 family. Docs now explain Cursor pricing as a third lookup source; overrides run after exact/prefix and before fuzzy, and the CLI labels them as Cursor.

Bug Fixes
- Exclude LiteLLM entries with prefix github_copilot/ to avoid zero-cost matches.
- Apply Cursor overrides after exact/prefix matches; skip with --provider; prefer upstream when present; handle version-normalized and provider-prefixed inputs and tier suffixes (e.g., gpt-5-3-codex → gpt-5.3-codex, openai/gpt-5.3-codex, -high); CLI shows “Cursor”; share compute_cost.
New Features
- Add Cursor pricing for gpt-5.3, gpt-5.3-codex, and gpt-5.3-codex-spark at $1.75/$14 per 1M tokens with $0.175 cache read; tests cover filtering, priority ordering, forced-source behavior, version normalization, provider-prefixed inputs, tier suffix stripping, and cost calculation; updated EN/KO/JA/ZH READMEs to document Cursor pricing and lookup order.

^{Written for commit 4989b0e. Summary will update on new commits.}

…overrides - Exclude github_copilot/ prefix from LiteLLM data (subscription-based $0.00 pricing poisons per-token cost estimation) - Inject hardcoded pricing overrides for gpt-5.3, gpt-5.3-codex, and gpt-5.3-codex-spark at $1.75/$14.00 per 1M tokens with $0.175 cache read (sourced from Cursor docs + llm-stats.com corroboration) - Overrides use or_insert_with to avoid clobbering future LiteLLM/OpenRouter entries once they add official API pricing - Add 6 unit tests covering filter, override injection, and lookup Closes #199, closes #172 Refs #204, #205

vercel · 2026-02-18T03:53:07Z

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

Project	Deployment	Actions	Updated (UTC)
tokscale	Ready	Preview, Comment	Feb 18, 2026 7:10am

cubic-dev-ai

No issues found across 2 files

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f249d32a63

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

packages/core/src/pricing/mod.rs

Overrides were injected as bare keys into the LiteLLM HashMap, which meant exact_match_litellm would return them before OpenRouter entries were checked. If OpenRouter added openai/gpt-5.3-codex with real API pricing, the hardcoded override would shadow it. Now overrides live in a separate fallback_overrides HashMap on PricingService, checked only AFTER both LiteLLM and OpenRouter return no match. Tests verify fallback yields to both LiteLLM bare keys and OpenRouter provider-prefixed entries.

cubic-dev-ai

1 issue found across 2 files (changes from recent commits).

Prompt for AI agents (all issues)


Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/core/src/pricing/mod.rs">

<violation number="1" location="packages/core/src/pricing/mod.rs:99">
P2: lookup_with_source now returns fallback pricing even when a specific source is forced, so `lookup_pricing(..., Some("litellm"))` can succeed with a "Fallback" source instead of returning an error. This violates the forced-source contract and can mislead callers.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

packages/core/src/pricing/mod.rs

…per, fix CLI label - Fallback overrides no longer fire when --provider is set; forced source miss correctly returns None (propagates error to CLI) - Extract compute_cost() shared between PricingLookup and PricingService to prevent cost formula drift - CLI now displays 'Fallback' (yellow) instead of mislabeling as OpenRouter - Add test_fallback_skipped_when_force_source_set

@keep

Fallback overrides are now checked after prefix matches but before fuzzy matching inside PricingLookup, preventing fuzzy from resolving gpt-5.3-codex to wrong models like gpt-5.1-codex-mini. Also restores doc comments with @keep annotations for non-obvious filtering and priority ordering logic. Closes #204, closes #205

cubic-dev-ai

1 issue found across 3 files (changes from recent commits).

Prompt for AI agents (all issues)


Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/core/src/pricing/lookup.rs">

<violation number="1" location="packages/core/src/pricing/lookup.rs:243">
P2: Fallback pricing only checks the raw model_id, so normalized variants (e.g., gpt-5-3-codex → gpt-5.3-codex) won’t match fallback even though version/model normalization is already applied for other sources. Apply the same normalization before fallback to keep behavior consistent.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

packages/core/src/pricing/lookup.rs

…n gap Rename all internal "fallback" naming to "cursor" since the data source is Cursor model pricing. Also fix a bug where version-normalized queries (e.g. gpt-5-3-codex → gpt-5.3-codex) would skip Cursor overrides and fall through to fuzzy matching, potentially resolving to wrong models.

…ntified tests openai/gpt-5.3-codex was resolving to perplexity/openai/gpt-4o-mini ($0.00) because exact_match_cursor did not strip provider prefixes. Now extracts the model part after "/" and checks Cursor overrides, matching the existing openrouter_model_part pattern. Added 3 tests for provider-prefix resolution, upstream precedence, and tier suffix stripping.

junhoyeo · 2026-02-18T07:01:16Z

Pricing Resolution Verification (49/49 models pass)

All models used on this machine resolve correctly after the fix. GPT-5.3 edge case coverage:

Query	Matched Key	Source	Input/1M	Output/1M
`gpt-5.3`	gpt-5.3	Cursor ✅	$1.75	$14.00
`gpt-5.3-codex`	gpt-5.3-codex	Cursor ✅	$1.75	$14.00
`gpt-5.3-codex-spark`	gpt-5.3-codex-spark	Cursor ✅	$1.75	$14.00
`gpt-5.3-codex-high`	gpt-5.3-codex	Cursor ✅	$1.75	$14.00
`gpt-5-3-codex`	gpt-5.3-codex	Cursor ✅	$1.75	$14.00
`openai/gpt-5.3-codex`	gpt-5.3-codex	Cursor ✅	$1.75	$14.00
`GPT-5.3-CODEX`	gpt-5.3-codex	Cursor ✅	$1.75	$14.00

Edge cases verified:

Tier suffix (-high, -xhigh, -low) → strips suffix, resolves to base model via Cursor
Version normalization (gpt-5-3-codex → gpt-5.3-codex) → normalizes before Cursor check
Provider prefix (openai/gpt-5.3-codex) → extracts model part, resolves via Cursor
Case insensitivity (GPT-5.3-CODEX) → lowercased before lookup
Upstream precedence → LiteLLM/OpenRouter exact/prefix matches always win over Cursor overrides

170 tests pass (cargo test), zero regressions on existing models.

Update lookup strategy (now 7 steps), pricing section, and features bullet across all 4 READMEs (EN, KO, JA, ZH-CN) to reflect the new Cursor pricing override layer for models not yet in LiteLLM/OpenRouter.

fix(pricing): GPT-5.3 Codex resolves to $0.00 — filter github_copilot + inject overrides Co-authored-by: Junho Yeo <i@junho.io>

vercel bot deployed to Preview February 18, 2026 03:53 View deployment

cubic-dev-ai bot reviewed Feb 18, 2026

View reviewed changes

chatgpt-codex-connector bot reviewed Feb 18, 2026

View reviewed changes

packages/core/src/pricing/mod.rs Outdated Show resolved Hide resolved

vercel bot deployed to Preview February 18, 2026 04:12 View deployment

cubic-dev-ai bot reviewed Feb 18, 2026

View reviewed changes

packages/core/src/pricing/mod.rs Outdated Show resolved Hide resolved

vercel bot deployed to Preview February 18, 2026 04:34 View deployment

vercel bot deployed to Preview February 18, 2026 06:08 View deployment

cubic-dev-ai bot reviewed Feb 18, 2026

View reviewed changes

packages/core/src/pricing/lookup.rs Outdated Show resolved Hide resolved

vercel bot deployed to Preview February 18, 2026 06:41 View deployment

vercel bot deployed to Preview February 18, 2026 07:00 View deployment

docs: document Cursor model pricing as third lookup source

4989b0e

Update lookup strategy (now 7 steps), pricing section, and features bullet across all 4 READMEs (EN, KO, JA, ZH-CN) to reflect the new Cursor pricing override layer for models not yet in LiteLLM/OpenRouter.

vercel bot deployed to Preview February 18, 2026 07:10 View deployment

junhoyeo merged commit 4be5f7c into main Feb 18, 2026
13 checks passed

junhoyeo deleted the fix/gpt-5.3-codex-pricing branch February 18, 2026 07:22

This was referenced Feb 18, 2026

gpt-5.3-codex pricing is wrong #172

Closed

Codex 5.3 no pricing estimation? #199

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(pricing): GPT-5.3 Codex resolves to $0.00 — filter github_copilot + inject overrides#207

fix(pricing): GPT-5.3 Codex resolves to $0.00 — filter github_copilot + inject overrides#207
junhoyeo merged 7 commits intomainfrom
fix/gpt-5.3-codex-pricing

junhoyeo commented Feb 18, 2026 •

edited by cubic-dev-ai bot

Loading

Uh oh!

vercel bot commented Feb 18, 2026 •

edited

Loading

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

Uh oh!

junhoyeo commented Feb 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

junhoyeo commented Feb 18, 2026 • edited by cubic-dev-ai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Root Cause

Test Results

Summary by cubic

Uh oh!

vercel bot commented Feb 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

junhoyeo commented Feb 18, 2026

Pricing Resolution Verification (49/49 models pass)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

junhoyeo commented Feb 18, 2026 •

edited by cubic-dev-ai bot

Loading

vercel bot commented Feb 18, 2026 •

edited

Loading