Summary
The "Cost by Model" breakdown appears to be pricing claude-sonnet-5 usage at the old Sonnet 4.x rate ($3 / MTok input, $15 / MTok output) instead of Sonnet 5's actual, permanent rate ($2 / MTok input, $10 / MTok output). This overstates Sonnet 5's estimated cost by roughly 50%.
Evidence
Row from the dashboard for claude-sonnet-5:
| Field |
Value |
| Turns |
8.4K |
| Input |
75.5K |
| Output |
4.96M |
| Cache Read |
1.99B |
| Cache Creation |
28.81M |
| Displayed Est. Cost |
$779.2297 |
Recomputing with Sonnet 5's official current pricing (per Anthropic's pricing page: $2/MTok input, $2.50/MTok 5-min cache write, $0.20/MTok cache read, $10/MTok output):
| Line item |
Calculation |
Cost |
| Input |
75,500 × $2 / 1,000,000 |
$0.15 |
| Output |
4,960,000 × $10 / 1,000,000 |
$49.60 |
| Cache read |
1,990,000,000 × $0.20 / 1,000,000 |
$398.00 |
| Cache creation (5m) |
28,810,000 × $2.50 / 1,000,000 |
$72.03 |
| Total |
|
≈ $519.78 |
That's a ~$259.45 overcount (~50%) on this row alone.
For comparison, if you apply the old Sonnet 4.x rate ($3/$3.75/$0.30/$15) to the same token counts, you get ≈ $779.67 — which matches the dashboard's displayed value almost exactly. This strongly suggests the cost calculator still has Sonnet 5 mapped to the previous Sonnet pricing tier.
Every other row in the same table (Opus 5, Haiku 4.5, Sonnet 4.6, Opus 4.6) reconciles correctly against their documented per-model rates, so this looks isolated to the claude-sonnet-5 price mapping specifically.
Expected behavior
claude-sonnet-5 usage should be billed/estimated at $2 / MTok input, $2.50 / MTok 5-min cache write, $4 / MTok 1-hour cache write, $0.20 / MTok cache read, $10 / MTok output — its actual, permanent rate (the planned increase to $3/$15 on Sept 1, 2026 was cancelled; $2/$10 is the standing price, not a temporary introductory rate).
Actual behavior
claude-sonnet-5 usage is being priced as if it were still $3 / MTok input, $15 / MTok output.
Suggested fix
Update the pricing table/lookup used by the cost calculator so claude-sonnet-5 (and any alias/model-string variants) maps to the $2/$10 rate rather than falling through to the legacy Sonnet 4.x rate.
Reference
https://platform.claude.com/docs/en/about-claude/pricing
Summary
The "Cost by Model" breakdown appears to be pricing
claude-sonnet-5usage at the old Sonnet 4.x rate ($3 / MTok input, $15 / MTok output) instead of Sonnet 5's actual, permanent rate ($2 / MTok input, $10 / MTok output). This overstates Sonnet 5's estimated cost by roughly 50%.Evidence
Row from the dashboard for
claude-sonnet-5:Recomputing with Sonnet 5's official current pricing (per Anthropic's pricing page: $2/MTok input, $2.50/MTok 5-min cache write, $0.20/MTok cache read, $10/MTok output):
That's a ~$259.45 overcount (~50%) on this row alone.
For comparison, if you apply the old Sonnet 4.x rate ($3/$3.75/$0.30/$15) to the same token counts, you get ≈ $779.67 — which matches the dashboard's displayed value almost exactly. This strongly suggests the cost calculator still has Sonnet 5 mapped to the previous Sonnet pricing tier.
Every other row in the same table (Opus 5, Haiku 4.5, Sonnet 4.6, Opus 4.6) reconciles correctly against their documented per-model rates, so this looks isolated to the
claude-sonnet-5price mapping specifically.Expected behavior
claude-sonnet-5usage should be billed/estimated at $2 / MTok input, $2.50 / MTok 5-min cache write, $4 / MTok 1-hour cache write, $0.20 / MTok cache read, $10 / MTok output — its actual, permanent rate (the planned increase to $3/$15 on Sept 1, 2026 was cancelled; $2/$10 is the standing price, not a temporary introductory rate).Actual behavior
claude-sonnet-5usage is being priced as if it were still $3 / MTok input, $15 / MTok output.Suggested fix
Update the pricing table/lookup used by the cost calculator so
claude-sonnet-5(and any alias/model-string variants) maps to the $2/$10 rate rather than falling through to the legacy Sonnet 4.x rate.Reference
https://platform.claude.com/docs/en/about-claude/pricing