diff --git a/src-tauri/src/acp/opencode_catalog.rs b/src-tauri/src/acp/opencode_catalog.rs
index 81b5bbed2..b3a65c0b8 100644
--- a/src-tauri/src/acp/opencode_catalog.rs
+++ b/src-tauri/src/acp/opencode_catalog.rs
@@ -53,6 +53,10 @@ pub struct CatalogModel {
     pub cost_in: Option<f64>,
     #[serde(default)]
     pub cost_out: Option<f64>,
+    #[serde(default)]
+    pub cost_cache_read: Option<f64>,
+    #[serde(default)]
+    pub cost_cache_write: Option<f64>,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -149,6 +153,14 @@ pub fn normalize_models_dev(raw: &str) -> Result<Vec<CatalogProvider>, AppComman
                         .get("cost")
                         .and_then(|v| v.get("output"))
                         .and_then(|v| v.as_f64()),
+                    cost_cache_read: m
+                        .get("cost")
+                        .and_then(|v| v.get("cache_read"))
+                        .and_then(|v| v.as_f64()),
+                    cost_cache_write: m
+                        .get("cost")
+                        .and_then(|v| v.get("cache_write"))
+                        .and_then(|v| v.as_f64()),
                 });
             }
         }
@@ -181,7 +193,13 @@ pub fn bundled_catalog() -> Vec<CatalogProvider> {
 }
 
 fn cache_path(data_dir: &Path) -> PathBuf {
-    data_dir.join("cache").join("opencode").join("models-dev.json")
+    // v2: cache_read / cache_write joined the slim shape. A new filename so
+    // an existing 24h cache is not served without those fields until it ages
+    // out, which would leave cache tokens unpriced on Token Usage.
+    data_dir
+        .join("cache")
+        .join("opencode")
+        .join("models-dev-v2.json")
 }
 
 fn read_cache(data_dir: &Path, require_fresh: bool) -> Option<Vec<CatalogProvider>> {
@@ -303,7 +321,17 @@ mod tests {
                         "reasoning": true,
                         "tool_call": true,
                         "limit": { "context": 128000, "output": 8192 },
-                        "cost": { "input": 1.5, "output": 6.0 }
+                        "cost": {
+                            "input": 1.5,
+                            "output": 6.0,
+                            "cache_read": 0.15,
+                            "cache_write": 1.875
+                        }
+                    },
+                    "demo-free": {
+                        "id": "demo-free",
+                        "name": "Demo Free",
+                        "cost": { "input": 0.0, "output": 0.0 }
                     }
                 }
             }
@@ -316,14 +344,27 @@ mod tests {
         assert_eq!(p.npm.as_deref(), Some("@ai-sdk/openai-compatible"));
         assert_eq!(p.env, vec!["DEMO_API_KEY".to_string()]);
         assert_eq!(p.auth_kind, "api");
-        assert_eq!(p.models.len(), 1);
-        let m = &p.models[0];
-        assert_eq!(m.id, "demo-large");
+        assert_eq!(p.models.len(), 2);
+        let m = p
+            .models
+            .iter()
+            .find(|m| m.id == "demo-large")
+            .expect("demo-large");
         assert!(m.reasoning);
         assert!(m.tool_call);
         assert_eq!(m.context, Some(128000));
         assert_eq!(m.cost_in, Some(1.5));
         assert_eq!(m.cost_out, Some(6.0));
+        assert_eq!(m.cost_cache_read, Some(0.15));
+        assert_eq!(m.cost_cache_write, Some(1.875));
+        let free = p
+            .models
+            .iter()
+            .find(|m| m.id == "demo-free")
+            .expect("demo-free");
+        assert_eq!(free.cost_in, Some(0.0));
+        assert_eq!(free.cost_cache_read, None);
+        assert_eq!(free.cost_cache_write, None);
     }
 
     #[test]
diff --git a/src/components/token-usage/token-usage-page.tsx b/src/components/token-usage/token-usage-page.tsx
index 841ff8e90..3724510d8 100644
--- a/src/components/token-usage/token-usage-page.tsx
+++ b/src/components/token-usage/token-usage-page.tsx
@@ -45,6 +45,7 @@ import { WorkbenchPageTitle } from "@/components/workbench/workbench-page-title"
 import { FolderAliasLabel } from "@/components/conversations/folder-alias-label"
 import { formatFolderLabelWithAlias } from "@/lib/folder-display"
 import {
+  opencodeProviderCatalog,
   tokenUsageFacets,
   tokenUsageReport,
   tokenUsageStatus,
@@ -65,6 +66,7 @@ import {
   foldBreakdown,
   formatDuration,
   formatTokensPrecise,
+  formatUsd,
   freshTokens,
   idleDays,
   localTzOffsetMinutes,
@@ -75,9 +77,16 @@ import {
   suggestBucket,
   type TokenUsageRangePreset,
 } from "@/lib/token-usage"
+import {
+  buildRateIndex,
+  estimateItemCost,
+  estimateReportCost,
+  resolveRate,
+} from "@/lib/model-api-rates"
 import { cn } from "@/lib/utils"
 import type {
   AgentType,
+  OpenCodeCatalogProvider,
   TokenUsageBucket,
   TokenUsageFacets,
   TokenUsageReport,
@@ -328,6 +337,7 @@ export function TokenUsagePage() {
   const [facets, setFacets] = useState<TokenUsageFacets | null>(null)
   const [status, setStatus] = useState<TokenUsageSyncStatus | null>(null)
   const [report, setReport] = useState<TokenUsageReport | null>(null)
+  const [catalog, setCatalog] = useState<OpenCodeCatalogProvider[] | null>(null)
   const [loading, setLoading] = useState(true)
   const [error, setError] = useState<string | null>(null)
   const [syncing, setSyncing] = useState(false)
@@ -432,6 +442,23 @@ export function TokenUsagePage() {
     void load()
   }, [load])
 
+  // Independent of the usage report: the same 24h models.dev catalog the
+  // OpenCode settings page already loads. Failures stay quiet and hide the
+  // dollar line instead of blocking the token counts.
+  useEffect(() => {
+    let cancelled = false
+    void opencodeProviderCatalog()
+      .then((list) => {
+        if (!cancelled) setCatalog(list)
+      })
+      .catch(() => {
+        if (!cancelled) setCatalog([])
+      })
+    return () => {
+      cancelled = true
+    }
+  }, [])
+
   // Latest-ref so the event subscription below is set up once, not torn down
   // and re-established on every filter change (`load`'s identity tracks the
   // filters). Same idiom as tasks-view-context.
@@ -561,6 +588,15 @@ export function TokenUsagePage() {
 
   const totals = report?.totals
   const cache = totals ? cacheHitRate(totals) : null
+  const rateIndex = useMemo(
+    () => (catalog && catalog.length > 0 ? buildRateIndex(catalog) : null),
+    [catalog]
+  )
+  const apiEstimate = useMemo(() => {
+    if (!report || !rateIndex) return null
+    const estimate = estimateReportCost(report.by_model, rateIndex)
+    return estimate.coverage > 0 && estimate.usd > 0 ? estimate : null
+  }, [report, rateIndex])
   const heat = useMemo(
     () => buildHeatMatrix(report?.heatmap ?? []),
     [report?.heatmap]
@@ -687,13 +723,23 @@ export function TokenUsagePage() {
     const { shown, other } = foldBreakdown(breakdownItems, BREAKDOWN_LIMIT)
     const share = (v: number) =>
       totals.total_tokens > 0 ? v / totals.total_tokens : null
-    const rows: RankedDatum[] = shown.map((it) => ({
-      key: it.key,
-      label: breakdownLabel(it.key, it.label),
-      value: it.total_tokens,
-      share: share(it.total_tokens),
-      hint: t("sessionsCount", { count: it.conversation_count }),
-    }))
+    const rows: RankedDatum[] = shown.map((it) => {
+      const sessions = t("sessionsCount", { count: it.conversation_count })
+      let hint = sessions
+      if (dim === "model" && rateIndex) {
+        const cost = estimateItemCost(it, resolveRate(rateIndex, it.key))
+        if (cost.usd != null && cost.usd > 0) {
+          hint = `${sessions} · ${formatUsd(cost.usd)}`
+        }
+      }
+      return {
+        key: it.key,
+        label: breakdownLabel(it.key, it.label),
+        value: it.total_tokens,
+        share: share(it.total_tokens),
+        hint,
+      }
+    })
     if (other) {
       rows.push({
         key: other.key,
@@ -705,7 +751,7 @@ export function TokenUsagePage() {
       })
     }
     return rows
-  }, [breakdownItems, breakdownLabel, totals, t])
+  }, [breakdownItems, breakdownLabel, totals, t, dim, rateIndex])
 
   const onBreakdownSelect = useCallback(
     (key: string) => {
@@ -1106,6 +1152,22 @@ export function TokenUsagePage() {
                                 />
                               )}
                             </div>
+                            {apiEstimate && (
+                              <div className="mt-2 text-sm text-muted-foreground">
+                                <span title={t("apiListEstimateHint")}>
+                                  {apiEstimate.coverage < 0.99
+                                    ? t("apiListEstimatePartial", {
+                                        amount: formatUsd(apiEstimate.usd),
+                                        percent: Math.floor(
+                                          apiEstimate.coverage * 100
+                                        ),
+                                      })
+                                    : t("apiListEstimate", {
+                                        amount: formatUsd(apiEstimate.usd),
+                                      })}
+                                </span>
+                              </div>
+                            )}
                             {archetype && totals.total_tokens > 0 && (
                               // The outer wrapper owns the spacing: mt-auto pins
                               // the strip to the card's bottom edge on wide
diff --git a/src/i18n/messages/ar.json b/src/i18n/messages/ar.json
index bdf81590d..bc58d2747 100644
--- a/src/i18n/messages/ar.json
+++ b/src/i18n/messages/ar.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "يقرأ codeg أعداد الرموز مباشرة من سجل كل وكيل. حدّث لتُحسب الجلسات الموجودة على هذا الجهاز.",
     "emptyAction": "احسب جلساتي",
     "loadFailed": "تعذّر تحميل بيانات الاستهلاك",
-    "truncatedNotice": "هذا النطاق واسع جدًا — الأرقام تغطي أحدث جزء منه فقط."
+    "truncatedNotice": "هذا النطاق واسع جدًا — الأرقام تغطي أحدث جزء منه فقط.",
+    "apiListEstimate": "≈ {amount} قائمة API",
+    "apiListEstimatePartial": "≈ {amount} قائمة API · {percent}٪ من الرموز",
+    "apiListEstimateHint": "أسعار قائمة API العامة من models.dev، وتشمل قراءة وكتابة الذاكرة المؤقتة. خطط الاشتراك تُحاسب بشكل مختلف."
   }
 }
diff --git a/src/i18n/messages/de.json b/src/i18n/messages/de.json
index 0c40db5d3..c453b07e8 100644
--- a/src/i18n/messages/de.json
+++ b/src/i18n/messages/de.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "Codeg liest die Token-Zahlen direkt aus dem Protokoll jedes Agenten. Aktualisiere, um zu zählen, was schon auf diesem Rechner liegt.",
     "emptyAction": "Meine Sitzungen zählen",
     "loadFailed": "Verbrauch konnte nicht geladen werden",
-    "truncatedNotice": "Dieser Zeitraum ist sehr groß — die Zahlen decken nur seinen jüngsten Abschnitt ab."
+    "truncatedNotice": "Dieser Zeitraum ist sehr groß — die Zahlen decken nur seinen jüngsten Abschnitt ab.",
+    "apiListEstimate": "≈ {amount} API-Liste",
+    "apiListEstimatePartial": "≈ {amount} API-Liste · {percent}% der Tokens",
+    "apiListEstimateHint": "Öffentliche API-Listenpreise von models.dev, inklusive Cache-Lesen und -Schreiben. Abo-Tarife rechnen anders ab."
   }
 }
diff --git a/src/i18n/messages/en.json b/src/i18n/messages/en.json
index 273af3aa0..2a71c74aa 100644
--- a/src/i18n/messages/en.json
+++ b/src/i18n/messages/en.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "Codeg reads token counts straight from each agent's own transcript. Run a refresh to count what's already on this machine.",
     "emptyAction": "Count my sessions",
     "loadFailed": "Could not load usage",
-    "truncatedNotice": "This range is very large — the numbers cover only the most recent slice of it."
+    "truncatedNotice": "This range is very large — the numbers cover only the most recent slice of it.",
+    "apiListEstimate": "≈ {amount} API list",
+    "apiListEstimatePartial": "≈ {amount} API list · {percent}% of tokens",
+    "apiListEstimateHint": "Public API list rates from models.dev, including cache reads and writes. Subscription plans bill differently."
   }
 }
diff --git a/src/i18n/messages/es.json b/src/i18n/messages/es.json
index 9948070a4..1cff8f36c 100644
--- a/src/i18n/messages/es.json
+++ b/src/i18n/messages/es.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "Codeg lee los tokens directamente de la transcripción de cada agente. Actualiza para contabilizar lo que ya hay en esta máquina.",
     "emptyAction": "Contabilizar mis sesiones",
     "loadFailed": "No se pudo cargar el uso",
-    "truncatedNotice": "Este periodo es muy amplio: las cifras cubren solo su tramo más reciente."
+    "truncatedNotice": "Este periodo es muy amplio: las cifras cubren solo su tramo más reciente.",
+    "apiListEstimate": "≈ {amount} tarifa API",
+    "apiListEstimatePartial": "≈ {amount} tarifa API · {percent}% de tokens",
+    "apiListEstimateHint": "Tarifas públicas de API de models.dev, con lecturas y escrituras de caché. Los planes de suscripción facturan distinto."
   }
 }
diff --git a/src/i18n/messages/fr.json b/src/i18n/messages/fr.json
index 9638508d6..bc8073b50 100644
--- a/src/i18n/messages/fr.json
+++ b/src/i18n/messages/fr.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "Codeg lit les tokens directement dans la transcription de chaque agent. Actualisez pour compter ce qui est déjà sur cette machine.",
     "emptyAction": "Compter mes sessions",
     "loadFailed": "Impossible de charger la consommation",
-    "truncatedNotice": "Cette période est très large — les chiffres ne couvrent que sa portion la plus récente."
+    "truncatedNotice": "Cette période est très large — les chiffres ne couvrent que sa portion la plus récente.",
+    "apiListEstimate": "≈ {amount} tarif API",
+    "apiListEstimatePartial": "≈ {amount} tarif API · {percent}% des jetons",
+    "apiListEstimateHint": "Tarifs publics API issus de models.dev, lectures et écritures de cache comprises. Les abonnements facturent autrement."
   }
 }
diff --git a/src/i18n/messages/ja.json b/src/i18n/messages/ja.json
index f44e8c88d..fee1acf19 100644
--- a/src/i18n/messages/ja.json
+++ b/src/i18n/messages/ja.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "codeg は各エージェント自身のセッション記録から直接トークン数を読み取ります。更新すると、このマシンにある記録を集計します。",
     "emptyAction": "セッションを集計",
     "loadFailed": "使用量を読み込めませんでした",
-    "truncatedNotice": "期間が非常に長いため、表示中の数値は直近の一部のみを対象としています。"
+    "truncatedNotice": "期間が非常に長いため、表示中の数値は直近の一部のみを対象としています。",
+    "apiListEstimate": "≈ {amount} APIリスト",
+    "apiListEstimatePartial": "≈ {amount} APIリスト · トークンの {percent}%",
+    "apiListEstimateHint": "models.dev の公開 API リスト料金です。キャッシュの読み書きを含みます。サブスクリプションの課金とは異なります。"
   }
 }
diff --git a/src/i18n/messages/ko.json b/src/i18n/messages/ko.json
index d8c404715..d82dd3347 100644
--- a/src/i18n/messages/ko.json
+++ b/src/i18n/messages/ko.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "codeg는 각 에이전트의 세션 기록에서 토큰 수를 직접 읽습니다. 새로 고침하면 이 컴퓨터에 있는 기록을 집계합니다.",
     "emptyAction": "내 세션 집계",
     "loadFailed": "사용량을 불러오지 못했습니다",
-    "truncatedNotice": "기간이 매우 길어 표시된 수치는 가장 최근 구간만 반영합니다."
+    "truncatedNotice": "기간이 매우 길어 표시된 수치는 가장 최근 구간만 반영합니다.",
+    "apiListEstimate": "≈ {amount} API 정가",
+    "apiListEstimatePartial": "≈ {amount} API 정가 · 토큰 {percent}%",
+    "apiListEstimateHint": "models.dev의 공개 API 정가이며 캐시 읽기/쓰기를 포함합니다. 구독 요금과는 다릅니다."
   }
 }
diff --git a/src/i18n/messages/pt.json b/src/i18n/messages/pt.json
index 5b31f9ead..39db7703d 100644
--- a/src/i18n/messages/pt.json
+++ b/src/i18n/messages/pt.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "O codeg lê os tokens direto da transcrição de cada agente. Atualize para contabilizar o que já existe nesta máquina.",
     "emptyAction": "Contabilizar minhas sessões",
     "loadFailed": "Não foi possível carregar o uso",
-    "truncatedNotice": "Este período é muito amplo — os números cobrem apenas o trecho mais recente."
+    "truncatedNotice": "Este período é muito amplo — os números cobrem apenas o trecho mais recente.",
+    "apiListEstimate": "≈ {amount} lista da API",
+    "apiListEstimatePartial": "≈ {amount} lista da API · {percent}% dos tokens",
+    "apiListEstimateHint": "Preços públicos da API em models.dev, incluindo leituras e escritas de cache. Planos de assinatura cobram diferente."
   }
 }
diff --git a/src/i18n/messages/zh-CN.json b/src/i18n/messages/zh-CN.json
index f6a2638c1..8b5590f16 100644
--- a/src/i18n/messages/zh-CN.json
+++ b/src/i18n/messages/zh-CN.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "codeg 直接从各智能体自己的会话记录里读取 Token 数。点一下刷新，把本机已有的会话统计进来。",
     "emptyAction": "统计我的会话",
     "loadFailed": "用量加载失败",
-    "truncatedNotice": "该范围过大 —— 下面的数字只覆盖了其中最近的一段。"
+    "truncatedNotice": "该范围过大 —— 下面的数字只覆盖了其中最近的一段。",
+    "apiListEstimate": "≈ {amount} API 标价",
+    "apiListEstimatePartial": "≈ {amount} API 标价 · {percent}% 的 token",
+    "apiListEstimateHint": "来自 models.dev 的公开 API 标价，含缓存读写。订阅套餐的计费方式不同。"
   }
 }
diff --git a/src/i18n/messages/zh-TW.json b/src/i18n/messages/zh-TW.json
index 13e7b6eae..7bb9da0b1 100644
--- a/src/i18n/messages/zh-TW.json
+++ b/src/i18n/messages/zh-TW.json
@@ -4951,6 +4951,9 @@
     "emptyHint": "codeg 直接從各智慧代理自己的工作階段紀錄讀取 Token 數。點一下重新整理，把本機已有的紀錄統計進來。",
     "emptyAction": "統計我的工作階段",
     "loadFailed": "用量載入失敗",
-    "truncatedNotice": "此範圍過大 —— 下面的數字只涵蓋其中最近的一段。"
+    "truncatedNotice": "此範圍過大 —— 下面的數字只涵蓋其中最近的一段。",
+    "apiListEstimate": "≈ {amount} API 標價",
+    "apiListEstimatePartial": "≈ {amount} API 標價 · {percent}% 的 token",
+    "apiListEstimateHint": "來自 models.dev 的公開 API 標價，含快取讀寫。訂閱方案的計費方式不同。"
   }
 }
diff --git a/src/lib/model-api-rates.test.ts b/src/lib/model-api-rates.test.ts
new file mode 100644
index 000000000..d565031bf
--- /dev/null
+++ b/src/lib/model-api-rates.test.ts
@@ -0,0 +1,269 @@
+import { describe, expect, it } from "vitest"
+
+import {
+  buildRateIndex,
+  estimateItemCost,
+  estimateReportCost,
+  pickRate,
+  resolveRate,
+  type ModelApiRate,
+} from "./model-api-rates"
+import type { OpenCodeCatalogProvider, TokenUsageBreakdownItem } from "./types"
+
+function model(
+  over: Partial<OpenCodeCatalogProvider["models"][number]> & { id: string }
+): OpenCodeCatalogProvider["models"][number] {
+  return {
+    name: over.id,
+    reasoning: false,
+    tool_call: true,
+    context: 200_000,
+    cost_in: 3,
+    cost_out: 15,
+    cost_cache_read: 0.3,
+    cost_cache_write: 3.75,
+    ...over,
+  }
+}
+
+function provider(
+  id: string,
+  models: OpenCodeCatalogProvider["models"],
+  over: Partial<OpenCodeCatalogProvider> = {}
+): OpenCodeCatalogProvider {
+  return {
+    id,
+    name: id,
+    npm: null,
+    env: [],
+    doc: null,
+    auth_kind: "api",
+    models,
+    ...over,
+  }
+}
+
+function item(
+  key: string,
+  over: Partial<TokenUsageBreakdownItem> = {}
+): TokenUsageBreakdownItem {
+  return {
+    key,
+    label: key,
+    input_tokens: 0,
+    output_tokens: 0,
+    cache_creation_tokens: 0,
+    cache_read_tokens: 0,
+    total_tokens: 0,
+    turn_count: 1,
+    conversation_count: 1,
+    ...over,
+  }
+}
+
+function rate(over: Partial<ModelApiRate> = {}): ModelApiRate {
+  return {
+    providerId: "anthropic",
+    id: "claude-sonnet-4-5",
+    name: "Claude Sonnet 4.5",
+    input: 3,
+    output: 15,
+    cacheRead: 0.3,
+    cacheWrite: 3.75,
+    ...over,
+  }
+}
+
+const CATALOG: OpenCodeCatalogProvider[] = [
+  provider("anthropic", [
+    model({
+      id: "claude-sonnet-4-5",
+      name: "Claude Sonnet 4.5",
+    }),
+  ]),
+  provider("openrouter", [
+    model({
+      id: "claude-sonnet-4-5",
+      name: "Claude Sonnet 4.5",
+      cost_in: 3.3,
+      cost_out: 16.5,
+      cost_cache_read: 0.33,
+      cost_cache_write: 4.125,
+    }),
+    model({
+      id: "gpt-5",
+      name: "GPT-5",
+      cost_in: 1.25,
+      cost_out: 10,
+      cost_cache_read: 0.125,
+      cost_cache_write: null,
+    }),
+  ]),
+  provider("openai", [
+    model({
+      id: "gpt-5",
+      name: "GPT-5",
+      cost_in: 1.25,
+      cost_out: 10,
+      cost_cache_read: 0.125,
+      cost_cache_write: null,
+    }),
+    model({
+      id: "gpt-5.3",
+      name: "GPT-5.3",
+      cost_in: 1.75,
+      cost_out: 14,
+      cost_cache_read: 0.175,
+      cost_cache_write: null,
+    }),
+  ]),
+  provider("xai", [
+    model({
+      id: "grok-4",
+      name: "Grok 4",
+      cost_in: 3,
+      cost_out: 15,
+      cost_cache_read: 0.75,
+      cost_cache_write: null,
+    }),
+  ]),
+]
+
+describe("resolveRate", () => {
+  const index = buildRateIndex(CATALOG)
+
+  it("matches an exact catalog id", () => {
+    expect(resolveRate(index, "claude-sonnet-4-5")?.providerId).toBe(
+      "anthropic"
+    )
+  })
+
+  it("strips a dated snapshot suffix", () => {
+    expect(resolveRate(index, "claude-sonnet-4-5-20250929")?.id).toBe(
+      "claude-sonnet-4-5"
+    )
+  })
+
+  it("honors an explicit provider prefix", () => {
+    expect(resolveRate(index, "openrouter/claude-sonnet-4-5")?.providerId).toBe(
+      "openrouter"
+    )
+  })
+
+  it("matches a display name", () => {
+    expect(resolveRate(index, "Claude Sonnet 4.5")?.id).toBe(
+      "claude-sonnet-4-5"
+    )
+  })
+
+  it("does not let gpt-5 steal gpt-5.3", () => {
+    expect(resolveRate(index, "gpt-5.3")?.id).toBe("gpt-5.3")
+    expect(resolveRate(index, "gpt-5")?.id).toBe("gpt-5")
+  })
+
+  it("accepts a numeric build suffix on an otherwise exact id", () => {
+    expect(resolveRate(index, "grok-4-0709")?.id).toBe("grok-4")
+  })
+
+  it("does not invent a rate for an unknown or folded key", () => {
+    expect(resolveRate(index, "mystery-model-9")).toBeNull()
+    expect(resolveRate(index, "__unknown__")).toBeNull()
+    expect(resolveRate(index, "__other__")).toBeNull()
+    expect(resolveRate(index, "")).toBeNull()
+  })
+})
+
+describe("pickRate", () => {
+  it("prefers the lab listing when two providers disagree", () => {
+    expect(
+      pickRate([
+        rate({ providerId: "openrouter", input: 9 }),
+        rate({ providerId: "anthropic", input: 3 }),
+      ])?.providerId
+    ).toBe("anthropic")
+  })
+
+  it("refuses to guess when two official listings disagree", () => {
+    expect(
+      pickRate([
+        rate({ providerId: "anthropic", input: 3 }),
+        rate({ providerId: "amazon-bedrock", input: 3.6 }),
+      ])
+    ).toBeNull()
+  })
+
+  it("collapses identical prices", () => {
+    expect(
+      pickRate([
+        rate({ providerId: "openrouter" }),
+        rate({ providerId: "anthropic" }),
+      ])?.providerId
+    ).toBe("anthropic")
+  })
+})
+
+describe("estimateItemCost", () => {
+  it("applies the four official list buckets", () => {
+    const cost = estimateItemCost(
+      item("claude-sonnet-4-5", {
+        input_tokens: 1_000_000,
+        output_tokens: 1_000_000,
+        cache_creation_tokens: 1_000_000,
+        cache_read_tokens: 1_000_000,
+      }),
+      rate()
+    )
+    expect(cost.usd).toBeCloseTo(3 + 15 + 3.75 + 0.3)
+    expect(cost.pricedTokens).toBe(4_000_000)
+    expect(cost.unpricedTokens).toBe(0)
+  })
+
+  it("leaves cache tokens unpriced when the catalog omitted that rate", () => {
+    const cost = estimateItemCost(
+      item("gpt-5", {
+        input_tokens: 1_000_000,
+        cache_read_tokens: 500_000,
+        cache_creation_tokens: 100_000,
+      }),
+      rate({ cacheRead: null, cacheWrite: null, input: 1.25, output: 10 })
+    )
+    expect(cost.usd).toBeCloseTo(1.25)
+    expect(cost.pricedTokens).toBe(1_000_000)
+    expect(cost.unpricedTokens).toBe(600_000)
+  })
+
+  it("returns no dollar amount when nothing could be priced", () => {
+    const cost = estimateItemCost(
+      item("mystery", { input_tokens: 10, output_tokens: 4 }),
+      null
+    )
+    expect(cost.usd).toBeNull()
+    expect(cost.unpricedTokens).toBe(14)
+  })
+})
+
+describe("estimateReportCost", () => {
+  const index = buildRateIndex(CATALOG)
+
+  it("sums only the priced models and reports coverage", () => {
+    const report = estimateReportCost(
+      [
+        item("claude-sonnet-4-5", {
+          input_tokens: 1_000_000,
+          total_tokens: 1_000_000,
+        }),
+        item("mystery-model", {
+          input_tokens: 250_000,
+          total_tokens: 250_000,
+        }),
+      ],
+      index
+    )
+    expect(report.usd).toBeCloseTo(3)
+    expect(report.pricedTokens).toBe(1_000_000)
+    expect(report.unpricedTokens).toBe(250_000)
+    expect(report.coverage).toBeCloseTo(0.8)
+    expect(report.pricedModels).toBe(1)
+    expect(report.unpricedModels).toBe(1)
+  })
+})
diff --git a/src/lib/model-api-rates.ts b/src/lib/model-api-rates.ts
new file mode 100644
index 000000000..2b22d0881
--- /dev/null
+++ b/src/lib/model-api-rates.ts
@@ -0,0 +1,298 @@
+import type {
+  OpenCodeCatalogProvider,
+  TokenUsageBreakdownItem,
+} from "@/lib/types"
+
+/**
+ * Public API list rates from the models.dev catalog Codeg already fetches.
+ *
+ * This is the official published per-million price (input, output, cache
+ * read, cache write). It is not what a subscription plan charged, and it is
+ * not a live remaining-quota figure.
+ */
+
+export interface ModelApiRate {
+  providerId: string
+  id: string
+  name: string
+  /** USD per 1M tokens. `null` means that bucket is unpublished. */
+  input: number | null
+  output: number | null
+  cacheRead: number | null
+  cacheWrite: number | null
+}
+
+export interface RateIndex {
+  byKey: Map<string, ModelApiRate[]>
+  byName: Map<string, ModelApiRate[]>
+  ids: string[]
+}
+
+export interface ItemCost {
+  /** `null` when no published rate applied to any token. */
+  usd: number | null
+  pricedTokens: number
+  unpricedTokens: number
+}
+
+export interface ReportCost {
+  usd: number
+  pricedTokens: number
+  unpricedTokens: number
+  /** priced / (priced + unpriced). 0 when nothing was recorded. */
+  coverage: number
+  pricedModels: number
+  unpricedModels: number
+}
+
+/** Prefer the lab's own listing when the same id is sold by many providers. */
+const PROVIDER_RANK: Record<string, number> = {
+  anthropic: 0,
+  openai: 1,
+  google: 2,
+  xai: 3,
+  groq: 4,
+  mistral: 5,
+  deepseek: 6,
+  "amazon-bedrock": 7,
+  azure: 8,
+}
+
+const DATE_SUFFIX = /-(?:\d{8}|\d{4}-\d{2}-\d{2})$/
+const PROVIDER_PREFIX = /^[a-z0-9][a-z0-9._-]*\//
+
+function finiteOrNull(n: number | null | undefined): number | null {
+  return typeof n === "number" && Number.isFinite(n) ? n : null
+}
+
+export function normalizeModelKey(raw: string): string {
+  return raw
+    .trim()
+    .toLowerCase()
+    .replace(/[_\s]+/g, "-")
+}
+
+function stripProvider(key: string): string {
+  return key.replace(PROVIDER_PREFIX, "")
+}
+
+function stripDate(key: string): string {
+  return key.replace(DATE_SUFFIX, "")
+}
+
+function lookupKeys(raw: string): string[] {
+  const n = normalizeModelKey(raw)
+  if (!n) return []
+  const noProv = stripProvider(n)
+  const keys = [n, noProv, stripDate(n), stripDate(noProv)]
+  return [...new Set(keys.filter(Boolean))]
+}
+
+function priceKey(rate: ModelApiRate): string {
+  return [rate.input, rate.output, rate.cacheRead, rate.cacheWrite].join("|")
+}
+
+function providerRank(id: string): number {
+  return PROVIDER_RANK[id] ?? 100
+}
+
+function push(
+  map: Map<string, ModelApiRate[]>,
+  key: string,
+  rate: ModelApiRate
+): void {
+  if (!key) return
+  const existing = map.get(key)
+  if (existing) existing.push(rate)
+  else map.set(key, [rate])
+}
+
+function toRate(
+  providerId: string,
+  model: OpenCodeCatalogProvider["models"][number]
+): ModelApiRate | null {
+  const input = finiteOrNull(model.cost_in)
+  const output = finiteOrNull(model.cost_out)
+  const cacheRead = finiteOrNull(model.cost_cache_read)
+  const cacheWrite = finiteOrNull(model.cost_cache_write)
+  if (
+    input == null &&
+    output == null &&
+    cacheRead == null &&
+    cacheWrite == null
+  ) {
+    return null
+  }
+  return {
+    providerId,
+    id: model.id,
+    name: model.name,
+    input,
+    output,
+    cacheRead,
+    cacheWrite,
+  }
+}
+
+/** Collapse several catalog hits to one rate, or none if the price is ambiguous. */
+export function pickRate(rates: ModelApiRate[]): ModelApiRate | null {
+  if (rates.length === 0) return null
+  if (rates.length === 1) return rates[0]
+
+  const uniquePrices = new Set(rates.map(priceKey))
+  const ranked = [...rates].sort(
+    (a, b) =>
+      providerRank(a.providerId) - providerRank(b.providerId) ||
+      a.providerId.localeCompare(b.providerId)
+  )
+  if (uniquePrices.size === 1) return ranked[0]
+
+  const official = ranked.filter((r) => providerRank(r.providerId) < 100)
+  if (official.length === 1) return official[0]
+  if (official.length > 1) {
+    const officialPrices = new Set(official.map(priceKey))
+    if (officialPrices.size === 1) return official[0]
+    return null
+  }
+  return null
+}
+
+export function buildRateIndex(
+  providers: OpenCodeCatalogProvider[]
+): RateIndex {
+  const byKey = new Map<string, ModelApiRate[]>()
+  const byName = new Map<string, ModelApiRate[]>()
+  const idSet = new Set<string>()
+
+  for (const provider of providers) {
+    for (const model of provider.models) {
+      const rate = toRate(provider.id, model)
+      if (!rate) continue
+      const id = normalizeModelKey(model.id)
+      const prefixed = `${normalizeModelKey(provider.id)}/${id}`
+      for (const key of lookupKeys(id)) push(byKey, key, rate)
+      push(byKey, prefixed, rate)
+      idSet.add(id)
+      const name = normalizeModelKey(model.name)
+      if (name.length >= 3) push(byName, name, rate)
+    }
+  }
+
+  return { byKey, byName, ids: [...idSet] }
+}
+
+export function resolveRate(
+  index: RateIndex,
+  raw: string | null | undefined
+): ModelApiRate | null {
+  if (!raw) return null
+  const trimmed = raw.trim()
+  if (!trimmed || trimmed === "__unknown__" || trimmed === "__other__") {
+    return null
+  }
+
+  for (const key of lookupKeys(trimmed)) {
+    const hit = pickRate(index.byKey.get(key) ?? [])
+    if (hit) return hit
+  }
+
+  for (const key of lookupKeys(trimmed)) {
+    const hit = pickRate(index.byName.get(key) ?? [])
+    if (hit) return hit
+  }
+
+  // `claude-sonnet-4-5-20250929` already matches via the date strip. This
+  // last pass only accepts a numeric build suffix (`grok-4-0709`) so
+  // `gpt-5` never steals `gpt-5.3` or `gpt-5-mini`.
+  const bare = stripProvider(normalizeModelKey(trimmed))
+  const prefixHits: ModelApiRate[] = []
+  for (const id of index.ids) {
+    if (!bare.startsWith(`${id}-`)) continue
+    if (!/^-\d{4,8}$/.test(bare.slice(id.length))) continue
+    prefixHits.push(...(index.byKey.get(id) ?? []))
+  }
+  return pickRate(prefixHits)
+}
+
+function priceBucket(
+  tokens: number,
+  perMillion: number | null
+): { usd: number; priced: number; unpriced: number } {
+  if (tokens <= 0) return { usd: 0, priced: 0, unpriced: 0 }
+  if (perMillion == null) return { usd: 0, priced: 0, unpriced: tokens }
+  return {
+    usd: (tokens / 1_000_000) * perMillion,
+    priced: tokens,
+    unpriced: 0,
+  }
+}
+
+export function estimateItemCost(
+  item: Pick<
+    TokenUsageBreakdownItem,
+    | "input_tokens"
+    | "output_tokens"
+    | "cache_creation_tokens"
+    | "cache_read_tokens"
+  >,
+  rate: ModelApiRate | null
+): ItemCost {
+  if (!rate) {
+    const unpriced =
+      item.input_tokens +
+      item.output_tokens +
+      item.cache_creation_tokens +
+      item.cache_read_tokens
+    return { usd: null, pricedTokens: 0, unpricedTokens: unpriced }
+  }
+
+  const input = priceBucket(item.input_tokens, rate.input)
+  const output = priceBucket(item.output_tokens, rate.output)
+  const cacheWrite = priceBucket(item.cache_creation_tokens, rate.cacheWrite)
+  const cacheRead = priceBucket(item.cache_read_tokens, rate.cacheRead)
+  const pricedTokens =
+    input.priced + output.priced + cacheWrite.priced + cacheRead.priced
+  const unpricedTokens =
+    input.unpriced + output.unpriced + cacheWrite.unpriced + cacheRead.unpriced
+  if (pricedTokens <= 0) {
+    return { usd: null, pricedTokens: 0, unpricedTokens }
+  }
+  return {
+    usd: input.usd + output.usd + cacheWrite.usd + cacheRead.usd,
+    pricedTokens,
+    unpricedTokens,
+  }
+}
+
+export function estimateReportCost(
+  items: TokenUsageBreakdownItem[],
+  index: RateIndex
+): ReportCost {
+  let usd = 0
+  let pricedTokens = 0
+  let unpricedTokens = 0
+  let pricedModels = 0
+  let unpricedModels = 0
+
+  for (const item of items) {
+    const cost = estimateItemCost(item, resolveRate(index, item.key))
+    if (cost.usd != null) {
+      usd += cost.usd
+      pricedModels += 1
+    } else if (cost.unpricedTokens > 0) {
+      unpricedModels += 1
+    }
+    pricedTokens += cost.pricedTokens
+    unpricedTokens += cost.unpricedTokens
+  }
+
+  const total = pricedTokens + unpricedTokens
+  return {
+    usd,
+    pricedTokens,
+    unpricedTokens,
+    coverage: total > 0 ? pricedTokens / total : 0,
+    pricedModels,
+    unpricedModels,
+  }
+}
diff --git a/src/lib/opencode-connect.test.ts b/src/lib/opencode-connect.test.ts
index c6c7cfb4a..49fadfd89 100644
--- a/src/lib/opencode-connect.test.ts
+++ b/src/lib/opencode-connect.test.ts
@@ -32,6 +32,8 @@ const CATALOG: OpenCodeCatalogProvider[] = [
         context: 400000,
         cost_in: 1,
         cost_out: 8,
+        cost_cache_read: null,
+        cost_cache_write: null,
       },
     ],
   },
@@ -51,6 +53,8 @@ const CATALOG: OpenCodeCatalogProvider[] = [
         context: 200000,
         cost_in: 0.5,
         cost_out: 2,
+        cost_cache_read: null,
+        cost_cache_write: null,
       },
     ],
   },
diff --git a/src/lib/token-usage.test.ts b/src/lib/token-usage.test.ts
index 6d646ed85..787900fdf 100644
--- a/src/lib/token-usage.test.ts
+++ b/src/lib/token-usage.test.ts
@@ -12,6 +12,7 @@ import {
   foldBreakdown,
   formatDuration,
   formatTokensPrecise,
+  formatUsd,
   freshTokens,
   fromLocalDayValue,
   idleDays,
@@ -562,6 +563,19 @@ describe("formatTokensPrecise", () => {
   })
 })
 
+describe("formatUsd", () => {
+  it("keeps tiny list prices readable instead of rounding them to $0.00", () => {
+    expect(formatUsd(0.0042)).toBe("$0.0042")
+    expect(formatUsd(0.01)).toBe("$0.01")
+  })
+
+  it("prints ordinary amounts with two decimals", () => {
+    expect(formatUsd(12.4)).toBe("$12.40")
+    expect(formatUsd(0)).toBe("$0")
+    expect(formatUsd(1500)).toBe("$1,500.00")
+  })
+})
+
 describe("formatDuration", () => {
   it("drops to the two largest useful units", () => {
     expect(formatDuration(0)).toBe("0s")
diff --git a/src/lib/token-usage.ts b/src/lib/token-usage.ts
index f11f96ba9..2e59a73af 100644
--- a/src/lib/token-usage.ts
+++ b/src/lib/token-usage.ts
@@ -498,6 +498,23 @@ export function formatTokensPrecise(n: number): string {
   return Math.round(n).toLocaleString()
 }
 
+/**
+ * USD list-price amounts for the usage hero. Always a dollar figure, never
+ * localized into another currency: the catalog rates are USD per 1M tokens.
+ */
+export function formatUsd(n: number): string {
+  if (!Number.isFinite(n) || n <= 0) return "$0"
+  if (n < 0.01) {
+    const trimmed = n.toFixed(4).replace(/0+$/, "").replace(/\.$/, "")
+    return `$${trimmed}`
+  }
+  if (n < 1000) return `$${n.toFixed(2)}`
+  return `$${n.toLocaleString("en-US", {
+    minimumFractionDigits: 2,
+    maximumFractionDigits: 2,
+  })}`
+}
+
 /** `1h 24m` / `3m 05s` / `12s` — generation time, never zero-padded hours. */
 export function formatDuration(ms: number): string {
   if (!Number.isFinite(ms) || ms <= 0) return "0s"
diff --git a/src/lib/types.ts b/src/lib/types.ts
index 8404ddc14..00d8f5d21 100644
--- a/src/lib/types.ts
+++ b/src/lib/types.ts
@@ -3486,6 +3486,8 @@ export interface OpenCodeCatalogModel {
   context: number | null
   cost_in: number | null
   cost_out: number | null
+  cost_cache_read: number | null
+  cost_cache_write: number | null
 }
 
 /**