devchain-cli 0.22.1 → 0.23.0

package/dist/server/modules/session-reader/data/pricing.json

@@ -211,6 +211,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "anthropic.claude-fable-5-1": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "cache_read_input_token_cost": 2.5e-7,
+        "cache_creation_input_token_cost": 0.0000125,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "global.anthropic.claude-fable-5": {
         "input_cost_per_token": 0.00001,
         "output_cost_per_token": 0.00005,
@@ -219,6 +227,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "global.anthropic.claude-fable-5-1": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "cache_read_input_token_cost": 2.5e-7,
+        "cache_creation_input_token_cost": 0.0000125,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "us.anthropic.claude-fable-5": {
         "input_cost_per_token": 0.000011,
         "output_cost_per_token": 0.000055,
@@ -227,6 +243,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "us.anthropic.claude-fable-5-1": {
+        "input_cost_per_token": 0.000011,
+        "output_cost_per_token": 0.000055,
+        "cache_read_input_token_cost": 2.75e-7,
+        "cache_creation_input_token_cost": 0.00001375,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "eu.anthropic.claude-fable-5": {
         "input_cost_per_token": 0.000011,
         "output_cost_per_token": 0.000055,
@@ -235,6 +259,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "eu.anthropic.claude-fable-5-1": {
+        "input_cost_per_token": 0.000011,
+        "output_cost_per_token": 0.000055,
+        "cache_read_input_token_cost": 2.75e-7,
+        "cache_creation_input_token_cost": 0.00001375,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "anthropic.claude-opus-5": {
         "input_cost_per_token": 0.000005,
         "output_cost_per_token": 0.000025,
@@ -588,6 +620,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "azure_ai/claude-fable-5-1": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "cache_read_input_token_cost": 2.5e-7,
+        "cache_creation_input_token_cost": 0.0000125,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "azure_ai/claude-opus-5": {
         "input_cost_per_token": 0.000005,
         "output_cost_per_token": 0.000025,
@@ -1454,6 +1494,7 @@
         "input_cost_per_token": 0.000005,
         "output_cost_per_token": 0.00003,
         "cache_read_input_token_cost": 5e-7,
+        "cache_creation_input_token_cost": 0.00000625,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1461,6 +1502,7 @@
         "input_cost_per_token": 0.000005,
         "output_cost_per_token": 0.00003,
         "cache_read_input_token_cost": 5e-7,
+        "cache_creation_input_token_cost": 0.00000625,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1468,6 +1510,7 @@
         "input_cost_per_token": 0.000002,
         "output_cost_per_token": 0.000012,
         "cache_read_input_token_cost": 2e-7,
+        "cache_creation_input_token_cost": 0.0000025,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1475,6 +1518,7 @@
         "input_cost_per_token": 2e-7,
         "output_cost_per_token": 0.0000012,
         "cache_read_input_token_cost": 2e-8,
+        "cache_creation_input_token_cost": 2.5e-7,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1482,6 +1526,7 @@
         "input_cost_per_token": 0.0000055,
         "output_cost_per_token": 0.000033,
         "cache_read_input_token_cost": 5.5e-7,
+        "cache_creation_input_token_cost": 0.000006875,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1489,6 +1534,7 @@
         "input_cost_per_token": 0.0000055,
         "output_cost_per_token": 0.000033,
         "cache_read_input_token_cost": 5.5e-7,
+        "cache_creation_input_token_cost": 0.000006875,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1496,6 +1542,7 @@
         "input_cost_per_token": 0.0000022,
         "output_cost_per_token": 0.0000132,
         "cache_read_input_token_cost": 2.2e-7,
+        "cache_creation_input_token_cost": 0.00000275,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1503,6 +1550,7 @@
         "input_cost_per_token": 2.2e-7,
         "output_cost_per_token": 0.00000132,
         "cache_read_input_token_cost": 2.2e-8,
+        "cache_creation_input_token_cost": 2.75e-7,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1510,6 +1558,7 @@
         "input_cost_per_token": 0.0000055,
         "output_cost_per_token": 0.000033,
         "cache_read_input_token_cost": 5.5e-7,
+        "cache_creation_input_token_cost": 0.000006875,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1517,6 +1566,7 @@
         "input_cost_per_token": 0.0000055,
         "output_cost_per_token": 0.000033,
         "cache_read_input_token_cost": 5.5e-7,
+        "cache_creation_input_token_cost": 0.000006875,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1524,6 +1574,7 @@
         "input_cost_per_token": 0.0000022,
         "output_cost_per_token": 0.0000132,
         "cache_read_input_token_cost": 2.2e-7,
+        "cache_creation_input_token_cost": 0.00000275,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -1531,6 +1582,7 @@
         "input_cost_per_token": 2.2e-7,
         "output_cost_per_token": 0.00000132,
         "cache_read_input_token_cost": 2.2e-8,
+        "cache_creation_input_token_cost": 2.75e-7,
         "max_input_tokens": 922000,
         "max_output_tokens": 128000
     },
@@ -2255,6 +2307,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "claude-fable-5-1": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "cache_read_input_token_cost": 2.5e-7,
+        "cache_creation_input_token_cost": 0.0000125,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "claude-opus-5": {
         "input_cost_per_token": 0.000005,
         "output_cost_per_token": 0.000025,
@@ -2323,96 +2383,168 @@
     "databricks/databricks-claude-3-7-sonnet": {
         "input_cost_per_token": 0.0000029999900000000002,
         "output_cost_per_token": 0.000015000020000000002,
+        "cache_read_input_token_cost": 3.0002e-7,
+        "cache_creation_input_token_cost": 0.00000374997,
         "max_input_tokens": 200000,
         "max_output_tokens": 128000
     },
+    "databricks/databricks-claude-fable-5": {
+        "input_cost_per_token": 0.00001000006,
+        "output_cost_per_token": 0.00005000002,
+        "cache_read_input_token_cost": 0.00000100002,
+        "cache_creation_input_token_cost": 0.00001250004,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "databricks/databricks-claude-haiku-4-5": {
         "input_cost_per_token": 0.00000100002,
         "output_cost_per_token": 0.00000500003,
+        "cache_read_input_token_cost": 1.0003e-7,
+        "cache_creation_input_token_cost": 0.00000124999,
         "max_input_tokens": 200000,
         "max_output_tokens": 64000
     },
     "databricks/databricks-claude-opus-4": {
         "input_cost_per_token": 0.000015000020000000002,
         "output_cost_per_token": 0.00007500003000000001,
+        "cache_read_input_token_cost": 0.00000150003,
+        "cache_creation_input_token_cost": 0.00001874999,
         "max_input_tokens": 200000,
         "max_output_tokens": 32000
     },
     "databricks/databricks-claude-opus-4-1": {
         "input_cost_per_token": 0.000015000020000000002,
         "output_cost_per_token": 0.00007500003000000001,
+        "cache_read_input_token_cost": 0.00000150003,
+        "cache_creation_input_token_cost": 0.00001874999,
         "max_input_tokens": 200000,
         "max_output_tokens": 32000
     },
     "databricks/databricks-claude-opus-4-5": {
         "input_cost_per_token": 0.00000500003,
         "output_cost_per_token": 0.000025000010000000002,
+        "cache_read_input_token_cost": 5.0001e-7,
+        "cache_creation_input_token_cost": 0.00000625002,
         "max_input_tokens": 200000,
         "max_output_tokens": 64000
     },
     "databricks/databricks-claude-opus-4-6": {
         "input_cost_per_token": 0.00000500003,
         "output_cost_per_token": 0.000025000010000000002,
+        "cache_read_input_token_cost": 5.0001e-7,
+        "cache_creation_input_token_cost": 0.00000625002,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "databricks/databricks-claude-opus-4-7": {
+        "input_cost_per_token": 0.00000500003,
+        "output_cost_per_token": 0.00002500001,
+        "cache_read_input_token_cost": 5.0001e-7,
+        "cache_creation_input_token_cost": 0.00000625002,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "databricks/databricks-claude-opus-4-8": {
+        "input_cost_per_token": 0.00000500003,
+        "output_cost_per_token": 0.00002500001,
+        "cache_read_input_token_cost": 5.0001e-7,
+        "cache_creation_input_token_cost": 0.00000625002,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "databricks/databricks-claude-opus-5": {
+        "input_cost_per_token": 0.00000500003,
+        "output_cost_per_token": 0.00002500001,
+        "cache_read_input_token_cost": 5.0001e-7,
+        "cache_creation_input_token_cost": 0.00000625002,
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
     "databricks/databricks-claude-sonnet-4": {
         "input_cost_per_token": 0.0000029999900000000002,
         "output_cost_per_token": 0.000015000020000000002,
+        "cache_read_input_token_cost": 3.0002e-7,
+        "cache_creation_input_token_cost": 0.00000374997,
         "max_input_tokens": 200000,
         "max_output_tokens": 64000
     },
     "databricks/databricks-claude-sonnet-4-1": {
         "input_cost_per_token": 0.0000029999900000000002,
         "output_cost_per_token": 0.000015000020000000002,
+        "cache_read_input_token_cost": 3.0002e-7,
+        "cache_creation_input_token_cost": 0.00000374997,
         "max_input_tokens": 200000,
         "max_output_tokens": 64000
     },
     "databricks/databricks-claude-sonnet-4-5": {
         "input_cost_per_token": 0.0000029999900000000002,
         "output_cost_per_token": 0.000015000020000000002,
+        "cache_read_input_token_cost": 3.0002e-7,
+        "cache_creation_input_token_cost": 0.00000374997,
         "max_input_tokens": 200000,
         "max_output_tokens": 64000
     },
     "databricks/databricks-claude-sonnet-4-6": {
         "input_cost_per_token": 0.0000029999900000000002,
         "output_cost_per_token": 0.000015000020000000002,
+        "cache_read_input_token_cost": 3.0002e-7,
+        "cache_creation_input_token_cost": 0.00000374997,
         "max_input_tokens": 1000000,
         "max_output_tokens": 64000
     },
+    "databricks/databricks-claude-sonnet-5": {
+        "input_cost_per_token": 0.00000299999,
+        "output_cost_per_token": 0.00001500002,
+        "cache_read_input_token_cost": 3.0002e-7,
+        "cache_creation_input_token_cost": 0.00000374997,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "databricks/databricks-gemini-2-5-flash": {
         "input_cost_per_token": 3.0001999999999996e-7,
         "output_cost_per_token": 0.00000249998,
+        "cache_read_input_token_cost": 3.0002e-8,
+        "cache_creation_input_token_cost": 3.0002e-7,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
     "databricks/databricks-gemini-2-5-pro": {
         "input_cost_per_token": 0.00000124999,
         "output_cost_per_token": 0.000009999990000000002,
+        "cache_read_input_token_cost": 1.24999e-7,
+        "cache_creation_input_token_cost": 0.00000124999,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
     "databricks/databricks-gemini-3-1-flash-lite": {
         "input_cost_per_token": 3.1248e-7,
         "output_cost_per_token": 0.00000187502,
+        "cache_read_input_token_cost": 3.122e-8,
+        "cache_creation_input_token_cost": 3.1248e-7,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
     "databricks/databricks-gemini-3-1-pro": {
         "input_cost_per_token": 0.00000249998,
         "output_cost_per_token": 0.000015000020000000002,
+        "cache_read_input_token_cost": 2.4997e-7,
+        "cache_creation_input_token_cost": 0.00000249998,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
     "databricks/databricks-gemini-3-flash": {
         "input_cost_per_token": 6.2503e-7,
         "output_cost_per_token": 0.00000374997,
+        "cache_read_input_token_cost": 6.251e-8,
+        "cache_creation_input_token_cost": 6.2503e-7,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
     "databricks/databricks-gemini-3-pro": {
         "input_cost_per_token": 0.00000249998,
         "output_cost_per_token": 0.000015000020000000002,
+        "cache_read_input_token_cost": 2.4997e-7,
+        "cache_creation_input_token_cost": 0.00000249998,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
@@ -2455,31 +2587,31 @@
     },
     "deepinfra/google/gemma-3-12b-it": {
         "input_cost_per_token": 5e-8,
-        "output_cost_per_token": 1e-7,
+        "output_cost_per_token": 1.5e-7,
         "max_input_tokens": 131072,
         "max_output_tokens": 131072
     },
     "deepinfra/google/gemma-3-27b-it": {
-        "input_cost_per_token": 9e-8,
+        "input_cost_per_token": 8e-8,
         "output_cost_per_token": 1.6e-7,
         "max_input_tokens": 131072,
         "max_output_tokens": 131072
     },
     "deepinfra/google/gemma-3-4b-it": {
-        "input_cost_per_token": 4e-8,
-        "output_cost_per_token": 8e-8,
+        "input_cost_per_token": 5e-8,
+        "output_cost_per_token": 1e-7,
         "max_input_tokens": 131072,
         "max_output_tokens": 131072
     },
     "deepinfra/openai/gpt-oss-120b": {
-        "input_cost_per_token": 5e-8,
-        "output_cost_per_token": 4.5e-7,
+        "input_cost_per_token": 3.7e-8,
+        "output_cost_per_token": 1.7e-7,
         "max_input_tokens": 131072,
         "max_output_tokens": 131072
     },
     "deepinfra/openai/gpt-oss-20b": {
-        "input_cost_per_token": 4e-8,
-        "output_cost_per_token": 1.5e-7,
+        "input_cost_per_token": 3e-8,
+        "output_cost_per_token": 1.4e-7,
         "max_input_tokens": 131072,
         "max_output_tokens": 131072
     },
@@ -2726,19 +2858,25 @@
     "gemini-2.5-flash-preview-09-2025": {
         "input_cost_per_token": 3e-7,
         "output_cost_per_token": 0.0000025,
-        "cache_read_input_token_cost": 7.5e-8,
+        "cache_read_input_token_cost": 3e-8,
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65535
+    },
+    "gemini-live-2.5-flash-native-audio": {
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.000002,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
     "gemini-live-2.5-flash-preview-native-audio-09-2025": {
-        "input_cost_per_token": 3e-7,
+        "input_cost_per_token": 5e-7,
         "output_cost_per_token": 0.000002,
         "cache_read_input_token_cost": 7.5e-8,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
     "gemini/gemini-live-2.5-flash-preview-native-audio-09-2025": {
-        "input_cost_per_token": 3e-7,
+        "input_cost_per_token": 5e-7,
         "output_cost_per_token": 0.000002,
         "cache_read_input_token_cost": 7.5e-8,
         "max_input_tokens": 1048576,
@@ -2747,7 +2885,7 @@
     "gemini-2.5-flash-lite-preview-06-17": {
         "input_cost_per_token": 1e-7,
         "output_cost_per_token": 4e-7,
-        "cache_read_input_token_cost": 2.5e-8,
+        "cache_read_input_token_cost": 1e-8,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
@@ -2834,6 +2972,13 @@
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
+    "vertex_ai/gemini-3.8-flash": {
+        "input_cost_per_token": 7.5e-7,
+        "output_cost_per_token": 0.00000375,
+        "cache_read_input_token_cost": 7.5e-8,
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536
+    },
     "vertex_ai/gemini-3.1-pro-preview": {
         "input_cost_per_token": 0.000002,
         "output_cost_per_token": 0.000012,
@@ -2857,12 +3002,9 @@
         "max_output_tokens": 65536
     },
     "gemini-2.5-pro-preview-tts": {
-        "input_cost_per_token": 0.00000125,
-        "output_cost_per_token": 0.00001,
+        "input_cost_per_token": 0.000001,
+        "output_cost_per_token": 0.00002,
         "cache_read_input_token_cost": 1.25e-7,
-        "input_cost_per_token_above_200k_tokens": 0.0000025,
-        "output_cost_per_token_above_200k_tokens": 0.000015,
-        "cache_read_input_token_cost_above_200k_tokens": 2.5e-7,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
@@ -2999,6 +3141,12 @@
         "max_input_tokens": 65536,
         "max_output_tokens": 32768
     },
+    "gemini/nano-banana-pro-preview": {
+        "input_cost_per_token": 0.000002,
+        "output_cost_per_token": 0.000012,
+        "max_input_tokens": 131072,
+        "max_output_tokens": 32768
+    },
     "gemini/gemini-3.1-flash-image": {
         "input_cost_per_token": 5e-7,
         "output_cost_per_token": 0.000003,
@@ -3040,34 +3188,34 @@
     "gemini/gemini-2.5-flash-preview-09-2025": {
         "input_cost_per_token": 3e-7,
         "output_cost_per_token": 0.0000025,
-        "cache_read_input_token_cost": 7.5e-8,
+        "cache_read_input_token_cost": 3e-8,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
     "gemini/gemini-flash-latest": {
         "input_cost_per_token": 3e-7,
         "output_cost_per_token": 0.0000025,
-        "cache_read_input_token_cost": 7.5e-8,
+        "cache_read_input_token_cost": 3e-8,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
     "gemini/gemini-flash-lite-latest": {
         "input_cost_per_token": 1e-7,
         "output_cost_per_token": 4e-7,
-        "cache_read_input_token_cost": 2.5e-8,
+        "cache_read_input_token_cost": 1e-8,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
     "gemini/gemini-2.5-flash-lite-preview-06-17": {
         "input_cost_per_token": 1e-7,
         "output_cost_per_token": 4e-7,
-        "cache_read_input_token_cost": 2.5e-8,
+        "cache_read_input_token_cost": 1e-8,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
     "gemini/gemini-2.5-flash-preview-tts": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.00001
     },
     "gemini/gemini-2.5-pro": {
         "input_cost_per_token": 0.00000125,
@@ -3146,11 +3294,18 @@
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
+    "gemini/gemini-3.8-flash": {
+        "input_cost_per_token": 7.5e-7,
+        "output_cost_per_token": 0.00000375,
+        "cache_read_input_token_cost": 7.5e-8,
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536
+    },
     "gemini/gemini-omni-flash-preview": {
         "input_cost_per_token": 0.0000015,
         "output_cost_per_token": 0.000009,
-        "max_input_tokens": 1048576,
-        "max_output_tokens": 65535
+        "max_input_tokens": 131072,
+        "max_output_tokens": 65536
     },
     "gemini/gemini-3.1-pro-preview": {
         "input_cost_per_token": 0.000002,
@@ -3206,13 +3361,17 @@
         "max_input_tokens": 1048576,
         "max_output_tokens": 65536
     },
+    "gemini-3.8-flash": {
+        "input_cost_per_token": 7.5e-7,
+        "output_cost_per_token": 0.00000375,
+        "cache_read_input_token_cost": 7.5e-8,
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536
+    },
     "gemini/gemini-2.5-pro-preview-tts": {
-        "input_cost_per_token": 0.00000125,
-        "output_cost_per_token": 0.00001,
+        "input_cost_per_token": 0.000001,
+        "output_cost_per_token": 0.00002,
         "cache_read_input_token_cost": 1.25e-7,
-        "input_cost_per_token_above_200k_tokens": 0.0000025,
-        "output_cost_per_token_above_200k_tokens": 0.000015,
-        "cache_read_input_token_cost_above_200k_tokens": 2.5e-7,
         "max_input_tokens": 1048576,
         "max_output_tokens": 65535
     },
@@ -3244,6 +3403,18 @@
         "max_input_tokens": 131072,
         "max_output_tokens": 8192
     },
+    "gemini/gemma-4-26b-a4b-it": {
+        "input_cost_per_token": 0,
+        "output_cost_per_token": 0,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 32768
+    },
+    "gemini/gemma-4-31b-it": {
+        "input_cost_per_token": 0,
+        "output_cost_per_token": 0,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 32768
+    },
     "gemini/learnlm-1.5-pro-experimental": {
         "input_cost_per_token": 0,
         "output_cost_per_token": 0,
@@ -3764,6 +3935,14 @@
         "max_input_tokens": 272000,
         "max_output_tokens": 128000
     },
+    "gpt-6-astra": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "cache_read_input_token_cost": 0.000001,
+        "cache_creation_input_token_cost": 0.0000125,
+        "max_input_tokens": 922000,
+        "max_output_tokens": 128000
+    },
     "gpt-5.6": {
         "input_cost_per_token": 0.000004,
         "output_cost_per_token": 0.00002,
@@ -4007,10 +4186,10 @@
     },
     "gpt-realtime-2": {
         "input_cost_per_token": 0.000004,
-        "output_cost_per_token": 0.000016,
+        "output_cost_per_token": 0.000024,
         "cache_read_input_token_cost": 4e-7,
-        "max_input_tokens": 32000,
-        "max_output_tokens": 4096
+        "max_input_tokens": 128000,
+        "max_output_tokens": 32000
     },
     "gpt-realtime-2.1": {
         "input_cost_per_token": 0.000004,
@@ -4024,12 +4203,12 @@
         "output_cost_per_token": 0.0000024,
         "cache_read_input_token_cost": 6e-8,
         "max_input_tokens": 128000,
-        "max_output_tokens": 4096
+        "max_output_tokens": 32000
     },
     "gpt-realtime-mini": {
         "input_cost_per_token": 6e-7,
         "output_cost_per_token": 0.0000024,
-        "max_input_tokens": 128000,
+        "max_input_tokens": 32000,
         "max_output_tokens": 4096
     },
     "gpt-realtime-2025-08-28": {
@@ -4706,13 +4885,22 @@
     "together_ai/openai/gpt-oss-120b": {
         "input_cost_per_token": 1.5e-7,
         "output_cost_per_token": 6e-7,
-        "max_input_tokens": 131072,
-        "max_output_tokens": 131072
+        "max_input_tokens": 131072
     },
     "together_ai/openai/gpt-oss-20b": {
         "input_cost_per_token": 5e-8,
         "output_cost_per_token": 2e-7,
-        "max_input_tokens": 128000
+        "max_input_tokens": 131072
+    },
+    "together_ai/google/gemma-3n-E4B-it": {
+        "input_cost_per_token": 6e-8,
+        "output_cost_per_token": 1.2e-7,
+        "max_input_tokens": 32768
+    },
+    "together_ai/google/gemma-4-31B-it": {
+        "input_cost_per_token": 3.9e-7,
+        "output_cost_per_token": 9.7e-7,
+        "max_input_tokens": 262144
     },
     "us.anthropic.claude-3-5-haiku-20241022-v1:0": {
         "input_cost_per_token": 8e-7,
@@ -4810,6 +4998,22 @@
         "max_input_tokens": 200000,
         "max_output_tokens": 64000
     },
+    "us-gov.anthropic.claude-sonnet-5": {
+        "input_cost_per_token": 0.0000024,
+        "output_cost_per_token": 0.000012,
+        "cache_read_input_token_cost": 2.4e-7,
+        "cache_creation_input_token_cost": 0.000003,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "us-gov.anthropic.claude-opus-4-8": {
+        "input_cost_per_token": 0.000006,
+        "output_cost_per_token": 0.00003,
+        "cache_read_input_token_cost": 6e-7,
+        "cache_creation_input_token_cost": 0.0000075,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "au.anthropic.claude-haiku-4-5-20251001-v1:0": {
         "input_cost_per_token": 0.0000011,
         "output_cost_per_token": 0.0000055,
@@ -5318,6 +5522,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "vertex_ai/claude-fable-5-1": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "cache_read_input_token_cost": 2.5e-7,
+        "cache_creation_input_token_cost": 0.0000125,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "vertex_ai/claude-fable-5@default": {
         "input_cost_per_token": 0.00001,
         "output_cost_per_token": 0.00005,
@@ -5326,6 +5538,14 @@
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000
     },
+    "vertex_ai/claude-fable-5-1@default": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "cache_read_input_token_cost": 2.5e-7,
+        "cache_creation_input_token_cost": 0.0000125,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
     "vertex_ai/claude-opus-5": {
         "input_cost_per_token": 0.000005,
         "output_cost_per_token": 0.000025,
@@ -5508,14 +5728,14 @@
         "max_output_tokens": 32768
     },
     "wandb/openai/gpt-oss-120b": {
-        "input_cost_per_token": 0.015,
-        "output_cost_per_token": 0.06,
+        "input_cost_per_token": 3e-8,
+        "output_cost_per_token": 1.7e-7,
         "max_input_tokens": 131072,
         "max_output_tokens": 131072
     },
     "wandb/openai/gpt-oss-20b": {
-        "input_cost_per_token": 0.005,
-        "output_cost_per_token": 0.02,
+        "input_cost_per_token": 3e-8,
+        "output_cost_per_token": 1.3e-7,
         "max_input_tokens": 131072,
         "max_output_tokens": 131072
     },
@@ -5653,20 +5873,20 @@
         "max_output_tokens": 8192
     },
     "gemini-2.5-flash-native-audio-latest": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025,
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.000002,
         "max_input_tokens": 1048576,
         "max_output_tokens": 8192
     },
     "gemini-2.5-flash-native-audio-preview-09-2025": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025,
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.000002,
         "max_input_tokens": 1048576,
         "max_output_tokens": 8192
     },
     "gemini-2.5-flash-native-audio-preview-12-2025": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025,
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.000002,
         "max_input_tokens": 1048576,
         "max_output_tokens": 8192
     },
@@ -5677,20 +5897,20 @@
         "max_output_tokens": 65536
     },
     "gemini/gemini-2.5-flash-native-audio-latest": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025,
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.000002,
         "max_input_tokens": 1048576,
         "max_output_tokens": 8192
     },
     "gemini/gemini-2.5-flash-native-audio-preview-09-2025": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025,
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.000002,
         "max_input_tokens": 1048576,
         "max_output_tokens": 8192
     },
     "gemini/gemini-2.5-flash-native-audio-preview-12-2025": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025,
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.000002,
         "max_input_tokens": 1048576,
         "max_output_tokens": 8192
     },
@@ -5707,8 +5927,8 @@
         "max_output_tokens": 16384
     },
     "gemini-2.5-flash-preview-tts": {
-        "input_cost_per_token": 3e-7,
-        "output_cost_per_token": 0.0000025
+        "input_cost_per_token": 5e-7,
+        "output_cost_per_token": 0.00001
     },
     "gemini-flash-latest": {
         "input_cost_per_token": 3e-7,
@@ -5875,5 +6095,181 @@
     "gemini/gemini-3.5-live-translate-preview": {
         "input_cost_per_token": 0.0000035,
         "output_cost_per_token": 0.000021
+    },
+    "gemini/gemini-3.5-transcribe": {
+        "input_cost_per_token": 0.000002,
+        "output_cost_per_token": 0.000012
+    },
+    "gemini/gemini-3.5-transcribe-live": {
+        "input_cost_per_token": 0.0000035,
+        "output_cost_per_token": 0.000021
+    },
+    "vertex_ai/gemini-3.5-transcribe-preview": {
+        "input_cost_per_token": 0.0000025,
+        "output_cost_per_token": 0.000012
+    },
+    "vertex_ai/gemini-3.5-transcribe-live-preview": {
+        "input_cost_per_token": 0.0000035,
+        "output_cost_per_token": 0.000021
+    },
+    "novita/google/gemma-4-26b-a4b-it": {
+        "input_cost_per_token": 1.3e-7,
+        "output_cost_per_token": 4.0000000000000003e-7,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 131072
+    },
+    "novita/google/gemma-4-31b-it": {
+        "input_cost_per_token": 1.4e-7,
+        "output_cost_per_token": 4.0000000000000003e-7,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 131072
+    },
+    "wandb/google/gemma-4-31B-it": {
+        "input_cost_per_token": 1e-7,
+        "output_cost_per_token": 3.4e-7,
+        "max_input_tokens": 262144
+    },
+    "deepinfra/openai/gpt-oss-120b-Turbo": {
+        "input_cost_per_token": 1.5e-7,
+        "output_cost_per_token": 6e-7,
+        "max_input_tokens": 131072
+    },
+    "deepinfra/google/gemma-4-31B-it-Ultra": {
+        "input_cost_per_token": 2.7e-7,
+        "output_cost_per_token": 7.6e-7,
+        "max_input_tokens": 131072
+    },
+    "deepinfra/anthropic/claude-opus-4-8": {
+        "input_cost_per_token": 0.000005,
+        "output_cost_per_token": 0.000025,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/anthropic/claude-sonnet-4-6": {
+        "input_cost_per_token": 0.000003,
+        "output_cost_per_token": 0.000015,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/google/gemini-3.5-flash": {
+        "input_cost_per_token": 0.0000015,
+        "output_cost_per_token": 0.000009,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/google/gemma-4-31B-it-turbo": {
+        "input_cost_per_token": 9e-8,
+        "output_cost_per_token": 3.4e-7,
+        "cache_read_input_token_cost": 5e-8,
+        "max_input_tokens": 262144
+    },
+    "deepinfra/anthropic/claude-opus-5": {
+        "input_cost_per_token": 0.000005,
+        "output_cost_per_token": 0.000025,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/google/gemini-3.1-flash-lite": {
+        "input_cost_per_token": 2.5e-7,
+        "output_cost_per_token": 0.0000015,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/google/gemini-3.7-flash": {
+        "input_cost_per_token": 7.5e-7,
+        "output_cost_per_token": 0.00000375,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/anthropic/claude-sonnet-5": {
+        "input_cost_per_token": 0.000002,
+        "output_cost_per_token": 0.00001,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/google/gemma-4-E4B-it": {
+        "input_cost_per_token": 2e-8,
+        "output_cost_per_token": 1e-7,
+        "max_input_tokens": 131072
+    },
+    "deepinfra/anthropic/claude-fable-5": {
+        "input_cost_per_token": 0.00001,
+        "output_cost_per_token": 0.00005,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/anthropic/claude-opus-4-7": {
+        "input_cost_per_token": 0.000005,
+        "output_cost_per_token": 0.000025,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/google/gemma-4-26B-A4B-it": {
+        "input_cost_per_token": 7e-8,
+        "output_cost_per_token": 3.4e-7,
+        "max_input_tokens": 262144
+    },
+    "deepinfra/google/gemini-3.1-pro": {
+        "input_cost_per_token": 0.000002,
+        "output_cost_per_token": 0.000012,
+        "max_input_tokens": 1000000
+    },
+    "deepinfra/anthropic/claude-haiku-4-5": {
+        "input_cost_per_token": 0.000001,
+        "output_cost_per_token": 0.000005,
+        "max_input_tokens": 200000
+    },
+    "deepinfra/openai/gpt-oss-120b-Ultra": {
+        "input_cost_per_token": 2e-7,
+        "output_cost_per_token": 9.5e-7,
+        "max_input_tokens": 131072
+    },
+    "deepinfra/google/gemma-4-31B-it": {
+        "input_cost_per_token": 1.3e-7,
+        "output_cost_per_token": 3.8e-7,
+        "max_input_tokens": 262144
+    },
+    "gemini/gemini-omni-1.1-flash": {
+        "input_cost_per_token": 0.0000015,
+        "output_cost_per_token": 0.000009,
+        "max_input_tokens": 131072,
+        "max_output_tokens": 65536
+    },
+    "bedrock/us-gov-west-1/anthropic.claude-sonnet-5": {
+        "input_cost_per_token": 0.0000024,
+        "output_cost_per_token": 0.000012,
+        "cache_read_input_token_cost": 2.4e-7,
+        "cache_creation_input_token_cost": 0.000003,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "bedrock/us-gov-west-1/anthropic.claude-opus-4-8": {
+        "input_cost_per_token": 0.000006,
+        "output_cost_per_token": 0.00003,
+        "cache_read_input_token_cost": 6e-7,
+        "cache_creation_input_token_cost": 0.0000075,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "bedrock/us-gov-east-1/anthropic.claude-sonnet-5": {
+        "input_cost_per_token": 0.0000024,
+        "output_cost_per_token": 0.000012,
+        "cache_read_input_token_cost": 2.4e-7,
+        "cache_creation_input_token_cost": 0.000003,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "bedrock/us-gov-east-1/anthropic.claude-opus-4-8": {
+        "input_cost_per_token": 0.000006,
+        "output_cost_per_token": 0.00003,
+        "cache_read_input_token_cost": 6e-7,
+        "cache_creation_input_token_cost": 0.0000075,
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000
+    },
+    "azure/us-gov/gpt-5.1": {
+        "input_cost_per_token": 0.00000171875,
+        "output_cost_per_token": 0.00001375,
+        "cache_read_input_token_cost": 1.71875e-7,
+        "max_input_tokens": 272000,
+        "max_output_tokens": 128000
+    },
+    "azure/us-gov/o3-mini": {
+        "input_cost_per_token": 0.000001513,
+        "output_cost_per_token": 0.00000605,
+        "cache_read_input_token_cost": 7.57e-7,
+        "max_input_tokens": 200000,
+        "max_output_tokens": 100000
     }
 }