[anascherif]

Problem

The Usage Dashboard calculates cost in calculate_cost (src-tauri/src/commands/usage.rs), and that function only knows two model families. It matches on substrings:

• any model containing opus-4 gets Opus 4's $15 / $75;
• any model containing sonnet-4 gets $3 / $15;
• every other model gets $0.

So, with current Claude Code models:

• Opus 4.5 to 4.8 ($5 / $25) are charged at the Opus 4 rate, 3x too much.
• Opus 5 / 5.5, Sonnet 5 / 5.5, Haiku 4.5 / 5.5 and Fable show $0.
• Cache writes are all charged at the 5-minute rate (1.25x input). Claude Code writes to the 1-hour cache, which is billed at 2x input. Its transcripts report the split in usage.cache_creation.ephemeral_1h_input_tokens.

On my own Claude Code history, the dashboard total was $1,168. With this change it is $736:

Model	Before	After
claude-opus-4-8	$1,168.20	$424.57
claude-opus-5-5	$0.00	$278.98
claude-fable-5-2	$0.00	$30.64
claude-haiku-4-5-20251001	$0.00	$0.70
claude-sonnet-5	$0.00	$0.49
claude-haiku-5-5	$0.00	$0.18

Approach

• model_pricing() reads the family and version out of the model ID and looks them up in a price table. It handles:

  • newer IDs (claude-opus-4-8);
  • dated snapshots (claude-sonnet-4-5-20250929);
  • the older naming (claude-3-5-haiku-20241022);
  • Bedrock IDs (us.anthropic.claude-opus-4-1-20250805-v1:0).

  The table lists input, output and cache-read prices. Cache-write prices follow Anthropic's multipliers: 1.25x input for 5 minutes, 2x for 1 hour.

• UsageData reads cache_creation.ephemeral_1h_input_tokens. calculate_cost charges that share at the 1-hour rate and the rest at the 5-minute rate. The 1-hour share is capped at cache_creation_input_tokens. Entries without the split are priced as before.

• Unknown models (for example <synthetic>) still cost $0, as before.

Testing

Added unit tests in usage.rs for:

• the price of each model generation;
• unknown models;
• 1-hour, mixed and 5-minute cache writes;
• a missing split;
• a 1-hour count larger than the cache writes;
• a row with every kind of token.

I don't have the Tauri build dependencies on this machine. So I ran these tests (6 passed) and cargo clippy (clean) on calculate_cost, model_pricing and the tests copied into a standalone crate. rustfmt --check on usage.rs is clean.

Side effects and limitations

• The price table needs a new entry when Anthropic releases a new model, as the old constants did. Until then, a new model shows $0, as unknown models do today.
• Long-context (over 200K tokens) premium pricing is not modelled, as before.