fix(eval): update DeepSeek V4 Flash pricing to current published rates

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.

[Astro-Han]

Summary

Follow-up from the #2971 review (@M4n5ter's first non-blocking item): the static DeepSeek pricing was stale.

DeepSeek's published V4 Flash pricing is now time-of-day dependent — peak hours (01:00–04:00 and 06:00–10:00 UTC) bill $0.44/M cache-miss input, $0.014/M cache-hit input and $1.32/M output; all other hours bill half. The old flat table ($0.14 / $0.0028 / $0.28) matches neither band and understated real spend by roughly 2.4× on a mostly off-peak run.

The metering checkpoint records aggregate usage without request timestamps, so a flat table remains the only representable shape. This updates the table to the off-peak band and documents the normalization: reported cost is a normalized API-equivalent, not an invoice, and a run straddling peak hours bills up to 2×. Rates fetched 2026-08-17 from https://api-docs.deepseek.com/quick_start/pricing.

Verification

• node --test dist/__tests__/external-subject.test.js — 13/13 pass, including the billing-identity test that reads the table symbolically.
• npm run format:check — clean.
• The 2.4× understatement was reproduced against the three-arm edit-contract run's metered usage, apportioned peak/off-peak by attempt wall-clock windows.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code authored the change and this PR body; a human contributor reviews the final diff and owns the merge decision. The commit carries a Generated-by trailer.

Checklist

• Tests cover the change and fail without it — the billing-identity test reads the table symbolically; the rate values are data with no failing test
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: ec51eadb-24a2-4332-a6fa-abfc2b11a2f8

📥 Commits

Reviewing files that changed from the base of the PR and between 9c37641 and 51626cb.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 1 remains after this review.

---

📝 Walkthrough

Summary

This PR updates DeepSeek V4 Flash pricing to current published off-peak rates:

• $0.22 per million cache-miss input tokens
• $0.007 per million cache-hit input tokens
• $0.66 per million output tokens

Peak-hour rates are twice the off-peak rates. Metering records do not include request timestamps, so reported costs remain normalized API-equivalent estimates. They are not invoices. Runs across peak hours can incur up to twice the normalized amount.

Design

The PR extends the existing pricing table and metering path. It does not create a parallel path.

The change is small and coherent. The added documentation is necessary because the flat pricing table cannot apply UTC time-dependent rates without request timestamps.

No code or tests can be deleted or simplified without reducing pricing coverage or cost-reporting clarity.

Validation

The PR reports 13/13 tests passing and clean formatting checks. Direct check output was not provided, so the final status of required checks remains unverified.

Review-relevant risks

The pricing update changes user-visible DeepSeek V4 Flash cost estimates. Material changes in user-visible behavior require independent human review under repository policy.

No public contract, security, licensing, release, or governance effect was identified in the current diff.

The person performing the merge must review the final diff. A maintainer makes the final determination.

Walkthrough

DeepSeek V4 Flash metering now uses updated off-peak input, output, cache-read, and cache-write rates. The code documents peak-hour variation and timestamp-free cost normalization.

Changes

DeepSeek V4 Flash metering

Layer / File(s)	Summary
Update pricing rates and cost basis packages/eval/src/provider-metering.ts	The exported rates use 0.22 input, 0.66 output, 0.007 cache read, and 0.22 cache write USD per million tokens. Documentation covers peak UTC pricing and timestamp-free cost normalization.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk:🔵 Low · up to 51626

The PR replaces stale DeepSeek V4 Flash pricing with normalized off-peak rates, improving reported cost accuracy. It is mergeable with owner awareness that the existing billing test will not catch an incorrect rate value, so an independent pricing-value check remains a useful follow-up.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: updating DeepSeek V4 Flash pricing rates.
Description check	✅ Passed	The description includes the required summary, verification, AI-use declaration, checklist, and behavior-change assessment.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Ai Use Disclosure	✅ Passed	The description selects generative tooling, names Claude Code and its scope, and the sole PR commit has the valid matching trailer Generated-by: Claude Code.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/eval-deepseek-pricing-current

---

_{Comment @coderabbitai help to get the list of available commands.}

[Copilot]

Pull request overview

Updates Eval’s DeepSeek V4 Flash pricing constants to match currently published rates and documents the new peak/off-peak pricing model while keeping a flat table (since metering checkpoints lack per-request timestamps).

Changes:

• Document DeepSeek’s time-of-day (peak/off-peak) pricing and clarify that reported cost is a normalized API-equivalent, not an invoice.
• Update DEEPSEEK_V4_FLASH_COST to the off-peak band values (so true bills during peak windows can be up to 2× the reported normalized cost).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 0332dbab-7f85-42ac-b78a-c2bab0ed2442

📥 Commits

Reviewing files that changed from the base of the PR and between 48e95e1 and 9c37641.

📒 Files selected for processing (1)

• packages/eval/src/provider-metering.ts

Included review availability: Your plan includes up to 3 reviews per rolling hour; 2 remain after this review.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Astro-Han]

Self-merge fast path: one-file update of a static pricing table and its comment — low impact, trivially reversible, outside the protected areas. Required checks pass. Reviewed by @Astro-Han. Follow-up from #2971's non-blocking review item.