@bitkyc08/opencodex 2.10.0 → 2.10.1

package/src/usage/expected-prices.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 };
@@ -37,7 +40,7 @@ const KIMI_K27_CODE: Cost4 = { input: 0.95, output: 4, cacheRead: 0.19, cacheWri
 const KIMI_K27_CODE_HIGHSPEED: Cost4 = { input: 1.9, output: 8, cacheRead: 0.38, cacheWrite: 1.9 };
 const KIMI_K26: Cost4 = { input: 0.95, output: 4, cacheRead: 0.16, cacheWrite: 0.95 };
 const KIMI_K25: Cost4 = { input: 0.6, output: 3, cacheRead: 0.1, cacheWrite: 0.6 };
-const QWEN38_ROUTEWAY_TEMPORARY: Cost4 = { input: 1.5, output: 5, cacheRead: 0.15, cacheWrite: 0 };
+const QWEN38_MAX: Cost4 = { input: 2, output: 6, cacheRead: 0, cacheWrite: 0 };
 // Anthropic official list prices (USD / 1M tokens). Cache write uses the published 5-minute rate.
 const CLAUDE_SONNET_46: Cost4 = { input: 3, output: 15, cacheRead: 0.3, cacheWrite: 3.75 };
 const CLAUDE_OPUS_46: Cost4 = { input: 5, output: 25, cacheRead: 0.5, cacheWrite: 6.25 };
@@ -50,13 +53,19 @@ 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.
 const KIMI_PRICING = "https://platform.kimi.ai/docs/pricing (official table; cacheWrite derived = input, Kimi auto-cache has no write billing)";
-// TEMPORARY proxy only: Routeway's reseller API rate is not Alibaba Token Plan billing.
-// Replace these overlays when Alibaba publishes an official qwen3.8-max-preview token rate.
-const QWEN38_ROUTEWAY_PRICING = "https://routeway.ai/models/qwen3.8-max-preview (temporary reseller proxy; NOT Alibaba Token Plan billing; cacheWrite unpublished -> 0)";
+// 260804: Qwen3.8-Max shipped as a stable model and Qwen published a per-token rate, which
+// is the exit condition the previous Routeway reseller overlay named. Two caveats are
+// deliberately in the source string rather than dropped: the figure comes from Qwen's own
+// release announcement, NOT from an Alibaba Model Studio billing table (which still lists
+// qwen3.7-max and qwen3-max but has no qwen3.8-max row), and no cache rate is published
+// anywhere. Cache stays 0 rather than inheriting the reseller's 0.15 — a reseller number
+// under a vendor-price label would be a wrong value wearing a verified badge.
+const QWEN38_MAX_PRICING = "https://qwen.ai/blog?id=qwen3.8 (Qwen release announcement; no Model Studio billing row yet; cache rates unpublished -> 0)";
 
 export const EXPECTED_PRICE_OVERLAYS: readonly ExpectedPriceOverlay[] = [
   // claude-opus-5 is exposed by three providers but absent from the jawcode bundle, so
@@ -77,6 +86,14 @@ 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" },
@@ -120,10 +137,10 @@ export const EXPECTED_PRICE_OVERLAYS: readonly ExpectedPriceOverlay[] = [
   { provider: "kimi-code", modelId: "kimi-k2.6", cost4: KIMI_K26, source: KIMI_PRICING, verifiedAt: "2026-07-20", status: "verified-derived" },
   { provider: "kimi-code", modelId: "kimi-k2.5", cost4: KIMI_K25, source: KIMI_PRICING, verifiedAt: "2026-07-20", status: "verified-derived" },
   { provider: "kimi-code", modelId: "kimi-for-coding", cost4: KIMI_K27_CODE, source: `derived: kimi-k2.7-code ${KIMI_PRICING}`, verifiedAt: "2026-07-20", status: "verified-derived" },
-  // Alibaba has not published a per-token Token Plan rate yet. Use Routeway's
-  // independently published reseller rate temporarily and keep estimates derived.
-  { provider: "alibaba-token-plan", modelId: "qwen3.8-max-preview", cost4: QWEN38_ROUTEWAY_TEMPORARY, source: QWEN38_ROUTEWAY_PRICING, verifiedAt: "2026-07-22", status: "verified-derived" },
-  { provider: "alibaba-token-plan-intl", modelId: "qwen3.8-max-preview", cost4: QWEN38_ROUTEWAY_TEMPORARY, source: QWEN38_ROUTEWAY_PRICING, verifiedAt: "2026-07-22", status: "verified-derived" },
+  // Qwen3.8-Max: vendor-published input/output rate (verified). See QWEN38_MAX_PRICING
+  // for what that source does and does not cover.
+  { provider: "alibaba-token-plan", modelId: "qwen3.8-max", cost4: QWEN38_MAX, source: QWEN38_MAX_PRICING, verifiedAt: "2026-08-04", status: "verified" },
+  { provider: "alibaba-token-plan-intl", modelId: "qwen3.8-max", cost4: QWEN38_MAX, source: QWEN38_MAX_PRICING, verifiedAt: "2026-08-04", status: "verified" },
   // Cursor Auto router — Cursor's published fixed token price (verified).
   { provider: "cursor", modelId: "auto", cost4: { input: 1.25, output: 6, cacheRead: 0.25, cacheWrite: 1.25 }, source: "https://docs.cursor.com/account/pricing + https://cursor.com/blog/aug-2025-pricing", verifiedAt: "2026-07-20", status: "verified" },
 ];
@@ -151,8 +168,11 @@ export function findExpectedPriceOverlay(
  */
 export const PRIORITY_MULTIPLIERS: Readonly<Record<string, number>> = {
   "gpt-5.6-sol": 2,
-  "gpt-5.6-terra": 1.6,
-  "gpt-5.6-luna": 0.4,
+  // Post-price-cut Fast tables (https://openai.com/api-fast-mode/, 2026-08-05):
+  // Terra 4/24/0.40/5 and Luna 0.40/2.40/0.04/0.50 are both 2× the corrected
+  // standard tuples; the stale bases made these look like 1.6/0.4 (#907).
+  "gpt-5.6-terra": 2,
+  "gpt-5.6-luna": 2,
   "gpt-5.5": 2.5,
   "gpt-5.4-mini": 2,
   "gpt-5.4": 2,
@@ -162,3 +182,102 @@ 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;
+}

package/src/usage/log.ts

@@ -4,14 +4,20 @@ import { getConfigDir } from "../config";
 import { recordOwnedConfigPath } from "../lib/config-ownership";
 import { usageDisplayTotalTokens } from "./totals";
 import type { OcxUsage } from "../types";
+import { normalizeRouteDecisionTrace, type RouteDecisionTraceV1 } from "../routing/trace";
 
 export type UsageStatus = "reported" | "unreported" | "unsupported" | "estimated";
 
+/**
+ * Recovery kinds recorded per attempt in the usage log; the GUI renders localized labels
+ * for these wire values.
+ */
 export type AttemptRecoveryKind =
   | "transient-5xx"
   | "connection-reset"
   | "oauth-401"
   | "key-429"
+  | "rate-limit-429"
   | "anthropic-oauth-429"
   | "image-413";
 
@@ -35,7 +41,7 @@ export interface PersistedUsageAttempt {
   requestedEffort?: string;
   effectiveEffort?: string;
   reasoningWireField?: string;
-  reasoningWireValue?: string | number;
+  reasoningWireValue?: string | number | boolean;
 }
 
 export interface PersistedUsageEntry {
@@ -59,7 +65,7 @@ export interface PersistedUsageEntry {
   /** Adapter-normalized tier and exact upstream parameter emitted for this request. */
   effectiveEffort?: string;
   reasoningWireField?: string;
-  reasoningWireValue?: string | number;
+  reasoningWireValue?: string | number | boolean;
   requestedServiceTier?: string;
   requestedSpeedLabel?: string;
   configuredServiceTier?: string;
@@ -81,6 +87,12 @@ export interface PersistedUsageEntry {
   closeReason?: "terminal" | "client_cancel" | "non_stream" | "body_stall" | "body_overflow";
   /** Already redacted + capped at capture (request-log.ts redactSecretString().slice(0,500)). */
   upstreamError?: string;
+  /**
+   * Bounded route-decision trace (RI-01): why this provider/model/account was
+   * selected. Additive field; old rows without it parse unchanged. Never
+   * contains prompts, credentials, or hidden reasoning.
+   */
+  routeDecision?: RouteDecisionTraceV1;
 }
 
 const KNOWN_USAGE_SURFACES = new Set<NonNullable<PersistedUsageEntry["surface"]>>([
@@ -173,6 +185,7 @@ const ATTEMPT_RECOVERY_KINDS = new Set<AttemptRecoveryKind>([
   "connection-reset",
   "oauth-401",
   "key-429",
+  "rate-limit-429",
   "anthropic-oauth-429",
   "image-413",
 ]);
@@ -268,14 +281,29 @@ function normalizeUsageAttempt(raw: unknown): PersistedUsageAttempt | null {
     ...(typeof attempt.reasoningWireField === "string" && attempt.reasoningWireField
       ? { reasoningWireField: capMetadataString(attempt.reasoningWireField) }
       : {}),
-    ...(typeof attempt.reasoningWireValue === "string" && attempt.reasoningWireValue
-      ? { reasoningWireValue: capMetadataString(attempt.reasoningWireValue) }
-      : isNonNegativeFiniteNumber(attempt.reasoningWireValue)
-        ? { reasoningWireValue: attempt.reasoningWireValue }
-        : {}),
+    ...(isValidReasoningWireValue(attempt.reasoningWireField, attempt.reasoningWireValue)
+      ? typeof attempt.reasoningWireValue === "string"
+        ? { reasoningWireValue: capMetadataString(attempt.reasoningWireValue) }
+        : { reasoningWireValue: attempt.reasoningWireValue }
+      : {}),
   };
 }
 
+/**
+ * Pairing rule for reasoning diagnostics, shared with the live request-log capture path:
+ * a non-empty string, a non-negative finite number, or a boolean only for
+ * `reasoning.enabled`. The field name itself is validated separately at capture time;
+ * persisted rows may carry legacy field names, so this checks only the value shape.
+ */
+export function isValidReasoningWireValue(
+  wireField: unknown,
+  wireValue: unknown,
+): wireValue is string | number | boolean {
+  return (typeof wireValue === "string" && wireValue.length > 0)
+    || (typeof wireValue === "number" && Number.isFinite(wireValue) && wireValue >= 0)
+    || (wireField === "reasoning.enabled" && typeof wireValue === "boolean");
+}
+
 function normalizedAttempts(raw: unknown): PersistedUsageAttempt[] {
   if (!Array.isArray(raw)) return [];
   return raw.map(normalizeUsageAttempt)
@@ -294,6 +322,9 @@ export function normalizeUsageEntryForTest(entry: PersistedUsageEntry): Persiste
 
 function normalizeUsageEntry(entry: PersistedUsageEntry): PersistedUsageEntry {
   const attempts = normalizedAttempts(entry.attempts);
+  const routeDecision = entry.routeDecision
+    ? normalizeRouteDecisionTrace(entry.routeDecision)
+    : undefined;
   return {
     requestId: entry.requestId,
     timestamp: entry.timestamp,
@@ -323,11 +354,11 @@ function normalizeUsageEntry(entry: PersistedUsageEntry): PersistedUsageEntry {
     ...(typeof entry.reasoningWireField === "string" && entry.reasoningWireField
       ? { reasoningWireField: capMetadataString(entry.reasoningWireField) }
       : {}),
-    ...(typeof entry.reasoningWireValue === "string" && entry.reasoningWireValue
-      ? { reasoningWireValue: capMetadataString(entry.reasoningWireValue) }
-      : isNonNegativeFiniteNumber(entry.reasoningWireValue)
-        ? { reasoningWireValue: entry.reasoningWireValue }
-        : {}),
+    ...(isValidReasoningWireValue(entry.reasoningWireField, entry.reasoningWireValue)
+      ? typeof entry.reasoningWireValue === "string"
+        ? { reasoningWireValue: capMetadataString(entry.reasoningWireValue) }
+        : { reasoningWireValue: entry.reasoningWireValue }
+      : {}),
     ...(typeof entry.requestedServiceTier === "string" && entry.requestedServiceTier
       ? { requestedServiceTier: capMetadataString(entry.requestedServiceTier) }
       : {}),
@@ -359,6 +390,7 @@ function normalizeUsageEntry(entry: PersistedUsageEntry): PersistedUsageEntry {
     ...(entry.terminalStatus ? { terminalStatus: entry.terminalStatus } : {}),
     ...(entry.closeReason ? { closeReason: entry.closeReason } : {}),
     ...(entry.upstreamError ? { upstreamError: entry.upstreamError } : {}),
+    ...(routeDecision ? { routeDecision } : {}),
   };
 }
 
@@ -623,9 +655,11 @@ export function readRecentUsageEntries(limit: number): PersistedUsageEntry[] {
     fd = openSync(path, "r");
     const size = fstatSync(fd).size;
     if (size <= 0) return [];
-    // ~4 KiB/row budget with a floor; expand once if the window yields too few lines.
-    let windowBytes = Math.min(size, Math.max(64 * 1024, Math.ceil(limit) * 4 * 1024));
-    for (let attempt = 0; attempt < 2; attempt++) {
+    // Trace-sized rows (up to MAX_TRACE_BYTES, RI-01) need a larger per-row
+    // budget than the pre-trace ledger; keep expanding until the window covers
+    // the file start or the whole file so a restart never hydrates nothing.
+    let windowBytes = Math.min(size, Math.max(64 * 1024, Math.ceil(limit) * 20 * 1024));
+    while (true) {
       const start = Math.max(0, size - windowBytes);
       const buf = Buffer.alloc(size - start);
       readSync(fd, buf, 0, buf.length, start);
@@ -645,6 +679,7 @@ export function readRecentUsageEntries(limit: number): PersistedUsageEntry[] {
       // most recent N valid rows (not N physical lines minus corrupt ones).
       const entries = parseUsageLines(lines);
       if (entries.length >= limit || start === 0 || windowBytes >= size) return entries.slice(-limit);
+      if (windowBytes >= size) break;
       windowBytes = Math.min(size, windowBytes * 4);
     }
     return [];

package/src/usage/summary.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 });

package/src/vision/index.ts

@@ -184,6 +184,11 @@ export function resolveVisionBackend(
   return anthropicSidecar ? "anthropic" : "openai";
 }
 
+/** Native model used by the OpenAI vision helper, including its bounded default. */
+export function resolveOpenAiVisionModel(config: Pick<OcxConfig, "visionSidecar">): string {
+  return config.visionSidecar?.model ?? DEFAULT_VISION_MODEL;
+}
+
 /** A user/developer/toolResult message can carry images (toolResult: e.g. Codex view_image output). */
 function carriesImages(role: string): boolean {
   return role === "user" || role === "developer" || role === "toolResult";
@@ -249,7 +254,7 @@ export function planVisionSidecar(
   return {
     backend,
     forwardSidecar: openAiSidecar,
-    settings: { model: cfg.model ?? DEFAULT_VISION_MODEL, timeoutMs: cfg.timeoutMs ?? DEFAULT_TIMEOUT_MS },
+    settings: { model: resolveOpenAiVisionModel(config), timeoutMs: cfg.timeoutMs ?? DEFAULT_TIMEOUT_MS },
     maxDescriptionsPerTurn,
   };
 }