fix(usage): price long-context requests at the published long rate (#908)

Several vendors reprice the entire request once the prompt crosses a token
threshold, and a flat Cost4 could not express it — so every request billed at
the short rate, including the long ones, which are the expensive ones.

The threshold reads raw usage.inputTokens, not normalized billable input: a
280k prompt with a 200k cache read has 80k billable input and still crosses
OpenAI's 272k boundary. Deciding after normalization would have under-billed
exactly the cache-heavy long requests.

Long context and Fast are mutually exclusive, not composable. OpenAI does not
serve long context in Fast mode, so exclusivity keys on the response-confirmed
tier: a >272k request merely tagged priority was necessarily downgraded and
bills long. That needed tier provenance at all four estimator call sites
instead of the collapsed scalar.

Also adds base prices for the three -pro virtual aliases, which resolved to
null and rendered no cost estimate at all.

Fixes #908

Author: lidge-jun

src/server/management/shared.ts

@@ -51,7 +51,7 @@ import type { OcxClaudeCodeConfig, OcxClaudeDesktopProfile, OcxConfig, OcxCustom
 import type { DesktopProfileModel } from "../../claude/desktop-profile";
 import { drainAndShutdown } from "../lifecycle";
 import { filterRequestLogs, getRequestLogEntries, type RequestLogEntry } from "../request-log";
-import { estimateComboCost, estimateRequestCost, effectiveServiceTier, normalizeCostTokens, tokensPerSecond } from "../../usage/cost";
+import { estimateComboCost, estimateRequestCost, serviceTierContext, normalizeCostTokens, tokensPerSecond } from "../../usage/cost";
 import type { PersistedUsageAttempt } from "../../usage/log";
 import { isAllowedRequestOrigin, jsonResponse, providerManagementConfigError, publicProviderBaseUrl, safeConfigDTO } from "../auth-cors";
 import { applySystemEnvToggle } from "../system-env";
@@ -124,7 +124,7 @@ export function unavailableCostReason(entry: MetricSource): MetricUnavailableRea
 }
 
 export function costResult(entry: MetricSource): CostResult {
-  const tier = effectiveServiceTier(entry);
+  const tier = serviceTierContext(entry);
   const estimate = entry.attempts?.length
     ? estimateComboCost(entry.attempts, undefined, tier)
     : estimateRequestCost({ provider: entry.provider, model: entry.model, usage: entry.usage, usageStatus: entry.usageStatus, serviceTier: tier });

src/usage/cost.ts

@@ -30,6 +30,9 @@ export interface ExpectedPriceOverlay {
 }
 
 const GEMINI_31_PRO: Cost4 = { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 0 };
+const GPT56_SOL: Cost4 = { input: 5, output: 30, cacheRead: 0.5, cacheWrite: 6.25 };
+const GPT56_TERRA: Cost4 = { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 2.5 };
+const GPT56_LUNA: Cost4 = { input: 0.2, output: 1.2, cacheRead: 0.02, cacheWrite: 0.25 };
 const GEMINI_36_FLASH: Cost4 = { input: 1.5, output: 7.5, cacheRead: 0.15, cacheWrite: 0 };
 const MINIMAX_M21_HIGHSPEED: Cost4 = { input: 0.6, output: 2.4, cacheRead: 0.03, cacheWrite: 0.375 };
 const KIMI_K3: Cost4 = { input: 3, output: 15, cacheRead: 0.3, cacheWrite: 3 };
@@ -50,6 +53,7 @@ const ANTHROPIC_PRICING = "https://platform.claude.com/docs/en/about-claude/pric
 
 const GEMINI_PRICING = "https://ai.google.dev/gemini-api/docs/pricing (2026-07-22); cacheWrite=0: storage is billed per-hour, not per-token";
 const MINIMAX_PRICING = "https://platform.minimax.io/docs/guides/pricing-paygo";
+const OPENAI_GPT56_PRICING = "https://developers.openai.com/api/docs/pricing";
 const DEEPSEEK_PRICING = "https://api-docs.deepseek.com/quick_start/pricing-details-usd; V4 Flash alias transition scheduled 2026-07-24 — re-verify after";
 // Kimi official tables publish input/output/cache-hit only; cacheWrite is mapped to the
 // cache-miss input price (Kimi auto-caches with no separate write billing). 2026-07-20 re-verified.
@@ -77,6 +81,14 @@ export const EXPECTED_PRICE_OVERLAYS: readonly ExpectedPriceOverlay[] = [
   // base model's standard rate per the official Billing FAQ).
   { provider: "google-antigravity", modelId: "gemini-3.6-flash", cost4: GEMINI_36_FLASH, source: `collapsed base ID ${GEMINI_PRICING}`, verifiedAt: "2026-07-22", status: "verified" },
   { provider: "google-antigravity", modelId: "gemini-3.1-pro", cost4: GEMINI_31_PRO, source: `collapsed base ID ${GEMINI_PRICING}`, verifiedAt: "2026-07-22", status: "verified" },
+  // OpenAI GPT-5.6 `-pro` virtual selections. The virtual resolver keeps the SELECTED id in
+  // the usage log and records the wire model separately, and cost resolution deliberately
+  // does not fall back through resolvedModel — so without these rows every `-pro` request
+  // resolved to null and rendered no cost estimate at all (#908 audit, runtime-verified).
+  // Pro reasoning bills at the base model's published API rate; the suffix is an effort knob.
+  { provider: "openai-apikey", modelId: "gpt-5.6-sol-pro", cost4: GPT56_SOL, source: `collapsed base ID ${OPENAI_GPT56_PRICING}`, verifiedAt: "2026-08-03", status: "verified-derived" },
+  { provider: "openai-apikey", modelId: "gpt-5.6-terra-pro", cost4: GPT56_TERRA, source: `collapsed base ID ${OPENAI_GPT56_PRICING}`, verifiedAt: "2026-08-03", status: "verified-derived" },
+  { provider: "openai-apikey", modelId: "gpt-5.6-luna-pro", cost4: GPT56_LUNA, source: `collapsed base ID ${OPENAI_GPT56_PRICING}`, verifiedAt: "2026-08-03", status: "verified-derived" },
   { provider: "google-antigravity", modelId: "gemini-3.1-pro-low", cost4: GEMINI_31_PRO, source: `derived: gemini-3.1-pro (<=200k tier) ${GEMINI_PRICING}`, verifiedAt: "2026-07-20", status: "verified-derived" },
   { provider: "google-antigravity", modelId: "gemini-3.1-pro-high", cost4: GEMINI_31_PRO, source: `derived: gemini-3.1-pro (<=200k tier) ${GEMINI_PRICING}`, verifiedAt: "2026-07-20", status: "verified-derived" },
   { provider: "google-antigravity", modelId: "gemini-pro-agent", cost4: GEMINI_31_PRO, source: `wire id for gemini-3.1-pro high ${GEMINI_PRICING}`, verifiedAt: "2026-07-23", status: "verified-derived" },
@@ -162,3 +174,102 @@ export const PRIORITY_MULTIPLIERS: Readonly<Record<string, number>> = {
 export function resolvePriorityMultiplier(modelId: string): number {
   return PRIORITY_MULTIPLIERS[modelId] ?? 1;
 }
+
+/**
+ * Long-context pricing tiers (#908). Several vendors reprice the ENTIRE request
+ * once the prompt crosses a published input-token threshold, so a flat Cost4
+ * cannot express it.
+ *
+ * The threshold is measured on RAW `usage.inputTokens` (total prompt size,
+ * including cache reads/writes) — never on normalized billable input, which has
+ * already had cache tokens subtracted. A 280k prompt with a 200k cache read has
+ * 80k billable input and still crosses OpenAI's 272k boundary; deciding after
+ * normalization would under-bill exactly the cache-heavy long requests.
+ *
+ * Rules are exact provider+model matches. No case folding: the jawcode bundle
+ * carries BOTH `minimax-m3` and `MiniMax-M3` at different rates, so folding
+ * would select the wrong base row. No model-level fallback: routed resellers
+ * (Cursor, OpenRouter) share model slugs but price independently.
+ */
+export interface ContextTier {
+  provider: string;
+  modelId: string;
+  /** Long rates apply once raw input tokens pass this boundary. */
+  thresholdInputTokens: number;
+  /** true = `>=` threshold (xAI), false = `>` threshold (OpenAI, MiniMax). */
+  inclusive: boolean;
+  /** Per-field factor from the short rate to the published long rate. */
+  multiplier: Cost4;
+  source: string;
+  verifiedAt: string;
+}
+
+/**
+ * OpenAI GPT-5.6: "Prompts with >272K input tokens are priced at 2x input and
+ * 1.5x output for the full request." Cached input and cache writes also double,
+ * per the published short/long columns.
+ */
+const OPENAI_LONG_CONTEXT: Cost4 = { input: 2, output: 1.5, cacheRead: 2, cacheWrite: 2 };
+/** xAI and MiniMax double every rate uniformly past their thresholds. */
+const UNIFORM_DOUBLE: Cost4 = { input: 2, output: 2, cacheRead: 2, cacheWrite: 2 };
+
+const OPENAI_PRICING_DOC = "https://developers.openai.com/api/docs/pricing";
+const OPENAI_GPT56_CONTEXT_MODELS = [
+  "gpt-5.6-sol",
+  "gpt-5.6-terra",
+  "gpt-5.6-luna",
+  // Virtual `-pro` selections keep their own id in usage logs (the wire model is
+  // recorded separately), so they need their own rows or they silently skip the tier.
+  "gpt-5.6-sol-pro",
+  "gpt-5.6-terra-pro",
+  "gpt-5.6-luna-pro",
+];
+
+export const CONTEXT_TIERS: readonly ContextTier[] = [
+  ...["openai", "openai-apikey"].flatMap(provider =>
+    OPENAI_GPT56_CONTEXT_MODELS.map((modelId): ContextTier => ({
+      provider,
+      modelId,
+      thresholdInputTokens: 272_000,
+      inclusive: false,
+      multiplier: OPENAI_LONG_CONTEXT,
+      source: OPENAI_PRICING_DOC,
+      verifiedAt: "2026-08-03",
+    })),
+  ),
+  {
+    provider: "xai",
+    modelId: "grok-4.5",
+    thresholdInputTokens: 200_000,
+    inclusive: true,
+    multiplier: UNIFORM_DOUBLE,
+    source: "https://docs.x.ai/developers/pricing",
+    verifiedAt: "2026-08-03",
+  },
+  ...["minimax", "minimax-cn"].map((provider): ContextTier => ({
+    provider,
+    modelId: "MiniMax-M3",
+    thresholdInputTokens: 512_000,
+    inclusive: false,
+    multiplier: UNIFORM_DOUBLE,
+    source: "https://platform.minimax.io/docs/guides/pricing-paygo",
+    verifiedAt: "2026-08-03",
+  })),
+];
+
+/** Exact provider+model context-tier lookup. No fuzzy matching, no case folding. */
+export function findContextTier(
+  provider: string,
+  modelId: string,
+  tiers: readonly ContextTier[] = CONTEXT_TIERS,
+): ContextTier | undefined {
+  return tiers.find(tier => tier.provider === provider && tier.modelId === modelId);
+}
+
+/** Whether a raw input-token count crosses the tier's published boundary. */
+export function isLongContext(tier: ContextTier, rawInputTokens: number): boolean {
+  if (!Number.isFinite(rawInputTokens)) return false;
+  return tier.inclusive
+    ? rawInputTokens >= tier.thresholdInputTokens
+    : rawInputTokens > tier.thresholdInputTokens;
+}

src/usage/expected-prices.ts

@@ -2,7 +2,7 @@ import { baseProviderLabel } from "../providers/label";
 import { canonicalAntigravityUsageModel } from "../providers/antigravity-models";
 import { usageDisplayTotalTokens } from "./totals";
 import type { PersistedUsageEntry, UsageStatus } from "./log";
-import { estimateComboCost, estimateRequestCost, effectiveServiceTier } from "./cost";
+import { estimateComboCost, estimateRequestCost, serviceTierContext } from "./cost";
 
 export type UsageRange = "7d" | "30d" | "all";
 export type UsageSurface = "all" | "codex" | "claude" | "grok";
@@ -286,7 +286,7 @@ function addEstimatedCost(
     totals.unmeteredRequests += 1;
     return;
   }
-  const tier = effectiveServiceTier(entry);
+  const tier = serviceTierContext(entry);
   const estimate = entry.attempts?.length
     ? estimateComboCost(entry.attempts, undefined, tier)
     : estimateRequestCost({ provider: entry.provider, model: entry.model, usage: entry.usage, usageStatus: entry.usageStatus, serviceTier: tier });
@@ -415,7 +415,7 @@ function buildModels(entries: PersistedUsageEntry[], totalTokens: number): Usage
   }
   // Accumulate per-model estimated cost
   for (const entry of entries) {
-    const tier = effectiveServiceTier(entry);
+    const tier = serviceTierContext(entry);
     const estimate = entry.attempts?.length
       ? estimateComboCost(entry.attempts, undefined, tier)
       : estimateRequestCost({ provider: entry.provider, model: entry.model, usage: entry.usage, usageStatus: entry.usageStatus, serviceTier: tier });
@@ -524,7 +524,7 @@ function buildProviders(entries: PersistedUsageEntry[], totalTokens: number): Us
     }
   }
   for (const entry of entries) {
-    const tier = effectiveServiceTier(entry);
+    const tier = serviceTierContext(entry);
     const estimate = entry.attempts?.length
       ? estimateComboCost(entry.attempts, undefined, tier)
       : estimateRequestCost({ provider: entry.provider, model: entry.model, usage: entry.usage, usageStatus: entry.usageStatus, serviceTier: tier });