Fix $0.00 cost on OpenRouter image generation models

Image-gen models are billed per-image/per-request, not at the plain
per-token prompt/completion rates ModelInfo.Pricing captures — so
token-based cost calculation was always near-zero for them regardless
of actual spend. Now requests OpenRouter's usage.include=true, decodes
the actual billed usage.cost, and prefers it over calculated cost via
a new resolveCost() helper (covers the dedicated Images API, regular
chat completions, streaming, and the tool-calling loop alike).
This commit is contained in:
2026-08-11 10:55:52 +02:00
parent ef6265c1c5
commit 1e0e81b9bc
4 changed files with 64 additions and 12 deletions
+14 -8
View File
@@ -1150,8 +1150,7 @@ Don't narrate future actions ("Let me...") - just use the tools.
if let usage = response.usage {
messages[index].tokens = usage.completionTokens
if let model = selectedModel {
let hasPricing = model.pricing.prompt > 0 || model.pricing.completion > 0
let cost: Double? = hasPricing ? Self.calculateCost(usage: usage, pricing: model.pricing) : nil
let cost = Self.resolveCost(usage: usage, pricing: model.pricing)
messages[index].cost = cost
sessionStats.addMessage(inputTokens: usage.promptTokens, outputTokens: usage.completionTokens, cost: cost)
}
@@ -1214,8 +1213,7 @@ Don't narrate future actions ("Let me...") - just use the tools.
if let usage = totalTokens {
messages[index].tokens = usage.completionTokens
if let model = selectedModel {
let hasPricing = model.pricing.prompt > 0 || model.pricing.completion > 0
let cost: Double? = hasPricing ? Self.calculateCost(usage: usage, pricing: model.pricing) : nil
let cost = Self.resolveCost(usage: usage, pricing: model.pricing)
messages[index].cost = cost
sessionStats.addMessage(inputTokens: usage.promptTokens, outputTokens: usage.completionTokens, cost: cost)
}
@@ -1859,8 +1857,7 @@ Don't narrate future actions ("Let me...") - just use the tools.
if willAutoContinue && finalContent.isEmpty {
// Nothing worth showing yet still record usage/cost for this turn.
if let usage = totalUsage, let model = selectedModel {
let hasPricing = model.pricing.prompt > 0 || model.pricing.completion > 0
let cost: Double? = hasPricing ? Self.calculateCost(usage: usage, pricing: model.pricing) : nil
let cost = Self.resolveCost(usage: usage, pricing: model.pricing)
sessionStats.addMessage(
inputTokens: usage.promptTokens,
outputTokens: usage.completionTokens,
@@ -1884,8 +1881,7 @@ Don't narrate future actions ("Let me...") - just use the tools.
// Calculate cost
if let usage = totalUsage, let model = selectedModel {
let hasPricing = model.pricing.prompt > 0 || model.pricing.completion > 0
let cost: Double? = hasPricing ? Self.calculateCost(usage: usage, pricing: model.pricing) : nil
let cost = Self.resolveCost(usage: usage, pricing: model.pricing)
if let index = messages.lastIndex(where: { $0.id == assistantMessage.id }) {
messages[index].cost = cost
}
@@ -2604,6 +2600,16 @@ Don't narrate future actions ("Let me...") - just use the tools.
return inputCost + cacheReadCost + cacheWriteCost + outputCost
}
/// Resolves a response's cost, preferring the provider's actual billed amount
/// (`usage.rawCostUSD` e.g. OpenRouter's `usage.include` cost, needed for models priced
/// outside plain per-token rates like per-image generation) over token-based calculation.
/// Falls back to `nil` when neither the raw cost nor per-token pricing is available.
nonisolated static func resolveCost(usage: ChatResponse.Usage, pricing: ModelInfo.Pricing) -> Double? {
if let raw = usage.rawCostUSD { return raw }
guard pricing.prompt > 0 || pricing.completion > 0 else { return nil }
return calculateCost(usage: usage, pricing: pricing)
}
/// Summarize a chunk of messages into a concise summary
private func summarizeMessageChunk(_ messages: [Message]) async -> String? {
guard let provider = providerRegistry.getProvider(for: currentProvider),