diff --git a/src/common/constants/knownModels.ts b/src/common/constants/knownModels.ts
index 16d7fca029..e537359af8 100644
--- a/src/common/constants/knownModels.ts
+++ b/src/common/constants/knownModels.ts
@@ -104,7 +104,7 @@ const MODEL_DEFINITIONS = {
     tokenizerOverride: "openai/gpt-5",
   },
   // GPT-5.6 Terra - balanced everyday tier, released July 9, 2026.
-  // GPT-5.5-class quality at half the cost: $2.50/M input, $15/M output; 1.05M context.
+  // GPT-5.5-class quality at a fraction of the cost: $2/M input, $12/M output; 1.05M context.
   GPT_56_TERRA: {
     provider: "openai",
     providerModelId: "gpt-5.6-terra",
@@ -112,7 +112,7 @@ const MODEL_DEFINITIONS = {
     tokenizerOverride: "openai/gpt-5",
   },
   // GPT-5.6 Luna - fastest, most cost-efficient tier, released July 9, 2026.
-  // $1/M input, $6/M output; 1.05M context (GA model page; 400K was a stale launch value).
+  // $0.20/M input, $1.20/M output; 1.05M context (GA model page; 400K was a stale launch value).
   GPT_56_LUNA: {
     provider: "openai",
     providerModelId: "gpt-5.6-luna",
diff --git a/src/common/utils/tokens/displayUsage.test.ts b/src/common/utils/tokens/displayUsage.test.ts
index 1ee9f04b32..34f53769e0 100644
--- a/src/common/utils/tokens/displayUsage.test.ts
+++ b/src/common/utils/tokens/displayUsage.test.ts
@@ -243,12 +243,13 @@ describe("createDisplayUsage", () => {
     expect(result!.input.tokens).toBe(500);
     expect(result!.cached.tokens).toBe(100000);
     expect(result!.cacheCreate.tokens).toBe(5000);
-    // Luna base rates: $1/M input, $0.10/M cache read (0.1x), $1.25/M cache
-    // write (1.25x), $6/M output.
-    expect(result!.input.cost_usd).toBeCloseTo(0.0005);
-    expect(result!.cached.cost_usd).toBeCloseTo(0.01);
-    expect(result!.cacheCreate.cost_usd).toBeCloseTo(0.00625);
-    expect(result!.output.cost_usd).toBeCloseTo(0.006);
+    // Luna base rates: $0.20/M input, $0.02/M cache read (0.1x), $0.25/M cache
+    // write (1.25x), $1.20/M output. These land near 1e-3, so the default
+    // toBeCloseTo precision of 2 would accept any of them; pin it explicitly.
+    expect(result!.input.cost_usd).toBeCloseTo(0.0001, 10);
+    expect(result!.cached.cost_usd).toBeCloseTo(0.002, 10);
+    expect(result!.cacheCreate.cost_usd).toBeCloseTo(0.00125, 10);
+    expect(result!.output.cost_usd).toBeCloseTo(0.0012, 10);
   });
 
   describe("tiered long-context pricing", () => {
diff --git a/src/common/utils/tokens/modelStats.test.ts b/src/common/utils/tokens/modelStats.test.ts
index 8e234944a6..2f2f6876c1 100644
--- a/src/common/utils/tokens/modelStats.test.ts
+++ b/src/common/utils/tokens/modelStats.test.ts
@@ -42,8 +42,8 @@ describe("getModelStats", () => {
   test.each([
     // [model, input, output, cacheRead, cacheCreation]
     ["openai:gpt-5.6-sol", 0.000005, 0.00003, 0.0000005, 0.00000625],
-    ["openai:gpt-5.6-terra", 0.0000025, 0.000015, 0.00000025, 0.000003125],
-    ["openai:gpt-5.6-luna", 0.000001, 0.000006, 0.0000001, 0.00000125],
+    ["openai:gpt-5.6-terra", 0.000002, 0.000012, 0.0000002, 0.0000025],
+    ["openai:gpt-5.6-luna", 0.0000002, 0.0000012, 0.00000002, 0.00000025],
   ] as const)(
     "resolves %s with the GA pricing and limits",
     (model, input, output, cacheRead, cacheCreation) => {
diff --git a/src/common/utils/tokens/models-extra.ts b/src/common/utils/tokens/models-extra.ts
index 2b2e78e5d0..b83d61928b 100644
--- a/src/common/utils/tokens/models-extra.ts
+++ b/src/common/utils/tokens/models-extra.ts
@@ -300,20 +300,21 @@ export const modelsExtra: Record<string, ModelData> = {
 
   // GPT-5.6 Terra - Released July 9, 2026 (balanced everyday tier).
   // GA docs: 1.05M context window, 128K max output, Feb 16 2026 cutoff (family).
-  // Base pricing: $2.50/M input, $15/M output, $0.25/M cached input; cache
-  // writes 1.25x the active input rate ($3.125/M base). Same 272K long-context
-  // tier as Sol (2x input / 1.5x output for the full request).
+  // Base pricing (OpenAI's July 30, 2026 price cut): $2/M input, $12/M output,
+  // $0.20/M cached input; cache writes 1.25x the active input rate ($2.50/M
+  // base). Same 272K long-context tier as Sol (2x input / 1.5x output for the
+  // full request).
   "gpt-5.6-terra": {
     max_input_tokens: 1050000,
     max_output_tokens: 128000,
-    input_cost_per_token: 0.0000025, // $2.50 per million input tokens (<272K prompt tokens)
-    input_cost_per_token_above_200k_tokens: 0.000005, // $5 per million input tokens (>272K)
-    output_cost_per_token: 0.000015, // $15 per million output tokens (<272K prompt tokens)
-    output_cost_per_token_above_200k_tokens: 0.0000225, // $22.50 per million output tokens (>272K)
-    cache_read_input_token_cost: 0.00000025, // $0.25 per million cached input tokens (<272K)
-    cache_read_input_token_cost_above_200k_tokens: 0.0000005, // $0.50 per million cached input tokens (>272K)
-    cache_creation_input_token_cost: 0.000003125, // $3.125 per million tokens (1.25x input)
-    cache_creation_input_token_cost_above_200k_tokens: 0.00000625, // $6.25 per million tokens (1.25x long-context input)
+    input_cost_per_token: 0.000002, // $2 per million input tokens (<272K prompt tokens)
+    input_cost_per_token_above_200k_tokens: 0.000004, // $4 per million input tokens (>272K)
+    output_cost_per_token: 0.000012, // $12 per million output tokens (<272K prompt tokens)
+    output_cost_per_token_above_200k_tokens: 0.000018, // $18 per million output tokens (>272K)
+    cache_read_input_token_cost: 0.0000002, // $0.20 per million cached input tokens (<272K)
+    cache_read_input_token_cost_above_200k_tokens: 0.0000004, // $0.40 per million cached input tokens (>272K)
+    cache_creation_input_token_cost: 0.0000025, // $2.50 per million tokens (1.25x input)
+    cache_creation_input_token_cost_above_200k_tokens: 0.000005, // $5 per million tokens (1.25x long-context input)
     tiered_pricing_threshold_tokens: 272000, // OpenAI's published boundary is 272K (field names say 200K)
     litellm_provider: "openai",
     mode: "chat",
@@ -327,20 +328,21 @@ export const modelsExtra: Record<string, ModelData> = {
   // GPT-5.6 Luna - Released July 9, 2026 (fastest, most cost-efficient tier).
   // GA model page: 1.05M context window (the 400K figure was a stale launch
   // value that caused premature compaction), 128K max output, Feb 16 2026 cutoff.
-  // Base pricing: $1/M input, $6/M output, $0.10/M cached input; cache writes
-  // 1.25x the active input rate ($1.25/M base). Same 272K long-context tier as
-  // Sol (2x input / 1.5x output for the full request).
+  // Base pricing (OpenAI's July 30, 2026 price cut): $0.20/M input, $1.20/M
+  // output, $0.02/M cached input; cache writes 1.25x the active input rate
+  // ($0.25/M base). Same 272K long-context tier as Sol (2x input / 1.5x output
+  // for the full request).
   "gpt-5.6-luna": {
     max_input_tokens: 1050000,
     max_output_tokens: 128000,
-    input_cost_per_token: 0.000001, // $1 per million input tokens (<272K prompt tokens)
-    input_cost_per_token_above_200k_tokens: 0.000002, // $2 per million input tokens (>272K)
-    output_cost_per_token: 0.000006, // $6 per million output tokens (<272K prompt tokens)
-    output_cost_per_token_above_200k_tokens: 0.000009, // $9 per million output tokens (>272K)
-    cache_read_input_token_cost: 0.0000001, // $0.10 per million cached input tokens (<272K)
-    cache_read_input_token_cost_above_200k_tokens: 0.0000002, // $0.20 per million cached input tokens (>272K)
-    cache_creation_input_token_cost: 0.00000125, // $1.25 per million tokens (1.25x input)
-    cache_creation_input_token_cost_above_200k_tokens: 0.0000025, // $2.50 per million tokens (1.25x long-context input)
+    input_cost_per_token: 0.0000002, // $0.20 per million input tokens (<272K prompt tokens)
+    input_cost_per_token_above_200k_tokens: 0.0000004, // $0.40 per million input tokens (>272K)
+    output_cost_per_token: 0.0000012, // $1.20 per million output tokens (<272K prompt tokens)
+    output_cost_per_token_above_200k_tokens: 0.0000018, // $1.80 per million output tokens (>272K)
+    cache_read_input_token_cost: 0.00000002, // $0.02 per million cached input tokens (<272K)
+    cache_read_input_token_cost_above_200k_tokens: 0.00000004, // $0.04 per million cached input tokens (>272K)
+    cache_creation_input_token_cost: 0.00000025, // $0.25 per million tokens (1.25x input)
+    cache_creation_input_token_cost_above_200k_tokens: 0.0000005, // $0.50 per million tokens (1.25x long-context input)
     tiered_pricing_threshold_tokens: 272000, // OpenAI's published boundary is 272K (field names say 200K)
     litellm_provider: "openai",
     mode: "chat",