Opus finding
Misleading pricing comment: claims Opus 4.7 'identical per-token prices to 4.6' but cites $5/$25 (current 4.6 published rates are $15/$75)
mediumdocs-gapmedium
- src/config.ts:27-33
The comment dates the rates to '2026-04' (future) and claims Opus 4.7 prices match Opus 4.6. Anthropic's publicly published Opus 4 / 4.x pricing has historically been $15 input / $75 output per 1M tokens, not $5/$25 (those are Sonnet-tier rates). Either the constants are wrong (undercounting cost 3×, defeating the entire 'Financial Safety' budget feature in lines 19-22) or the comment is misleading about which model these apply to. Given that MODELS.high is 'claude-opus-4-7' and the prompt calls itself Opus 4.7, but the budgeting math uses Sonnet-priced constants, sessions can blow through real spend ~3× faster than the budget guard reports. This is a financial-safety regression.
Recommendation
Verify Opus 4.7 pricing against Anthropic's docs and either (a) correct the constants, or (b) make pricing model-dependent (a map keyed by MODELS) so high/medium/low tiers use the correct per-token rates, and budget computation picks the right one. At minimum fix the comment to reflect reality.