From e696b33e0f9ee09230eedc8d7c59f376e5a96caa Mon Sep 17 00:00:00 2001 From: Aiden Cline Date: Mon, 8 Jun 2026 20:16:48 -0500 Subject: [PATCH 1/3] feat(google-vertex): add reasoning options --- .../models/claude-3-7-sonnet@20250219.toml | 7 ----- .../models/claude-haiku-4-5@20251001.toml | 1 + .../models/claude-opus-4-1@20250805.toml | 2 ++ .../models/claude-opus-4-5@20251101.toml | 1 + .../models/claude-opus-4-6@default.toml | 1 + .../models/claude-opus-4-7@default.toml | 1 + .../models/claude-opus-4-8@default.toml | 1 + .../models/claude-opus-4@20250514.toml | 2 ++ .../models/claude-sonnet-4-5@20250929.toml | 1 + .../models/claude-sonnet-4-6@default.toml | 1 + .../models/claude-sonnet-4@20250514.toml | 2 ++ .../models/claude-3-7-sonnet@20250219.toml | 10 ------- .../models/claude-haiku-4-5@20251001.toml | 1 + .../models/claude-opus-4-1@20250805.toml | 2 ++ .../models/claude-opus-4-5@20251101.toml | 1 + .../models/claude-opus-4-6@default.toml | 1 + .../models/claude-opus-4-7@default.toml | 1 + .../models/claude-opus-4-8@default.toml | 1 + .../models/claude-opus-4@20250514.toml | 2 ++ .../models/claude-sonnet-4-5@20250929.toml | 1 + .../models/claude-sonnet-4-6@default.toml | 1 + .../models/claude-sonnet-4@20250514.toml | 2 ++ .../gemini-2.5-flash-lite-preview-06-17.toml | 23 -------------- .../models/gemini-2.5-flash-lite.toml | 1 + .../gemini-2.5-flash-preview-09-2025.toml | 24 --------------- .../models/gemini-2.5-flash.toml | 1 + .../google-vertex/models/gemini-2.5-pro.toml | 1 + .../models/gemini-3-flash-preview.toml | 1 + .../models/gemini-3-pro-preview.toml | 30 ------------------- .../models/gemini-3.1-flash-lite-preview.toml | 1 + .../models/gemini-3.1-flash-lite.toml | 1 + .../gemini-3.1-pro-preview-customtools.toml | 1 + .../models/gemini-3.1-pro-preview.toml | 1 + .../models/gemini-3.5-flash.toml | 1 + .../models/gemini-flash-latest.toml | 24 --------------- .../models/gemini-flash-lite-latest.toml | 7 ----- 36 files changed, 35 insertions(+), 125 deletions(-) delete mode 100644 providers/google-vertex-anthropic/models/claude-3-7-sonnet@20250219.toml delete mode 100644 providers/google-vertex/models/claude-3-7-sonnet@20250219.toml delete mode 100644 providers/google-vertex/models/gemini-2.5-flash-lite-preview-06-17.toml delete mode 100644 providers/google-vertex/models/gemini-2.5-flash-preview-09-2025.toml delete mode 100644 providers/google-vertex/models/gemini-3-pro-preview.toml delete mode 100644 providers/google-vertex/models/gemini-flash-latest.toml delete mode 100644 providers/google-vertex/models/gemini-flash-lite-latest.toml diff --git a/providers/google-vertex-anthropic/models/claude-3-7-sonnet@20250219.toml b/providers/google-vertex-anthropic/models/claude-3-7-sonnet@20250219.toml deleted file mode 100644 index 56b0d0eca..000000000 --- a/providers/google-vertex-anthropic/models/claude-3-7-sonnet@20250219.toml +++ /dev/null @@ -1,7 +0,0 @@ -base_model = "anthropic/claude-3-7-sonnet-20250219" - -[cost] -input = 3 -output = 15 -cache_read = 0.3 -cache_write = 3.75 diff --git a/providers/google-vertex-anthropic/models/claude-haiku-4-5@20251001.toml b/providers/google-vertex-anthropic/models/claude-haiku-4-5@20251001.toml index 44cfe300a..2cc27f392 100644 --- a/providers/google-vertex-anthropic/models/claude-haiku-4-5@20251001.toml +++ b/providers/google-vertex-anthropic/models/claude-haiku-4-5@20251001.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-haiku-4-5-20251001" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] [cost] input = 1 diff --git a/providers/google-vertex-anthropic/models/claude-opus-4-1@20250805.toml b/providers/google-vertex-anthropic/models/claude-opus-4-1@20250805.toml index e06ff7bc0..89ead5537 100644 --- a/providers/google-vertex-anthropic/models/claude-opus-4-1@20250805.toml +++ b/providers/google-vertex-anthropic/models/claude-opus-4-1@20250805.toml @@ -1,4 +1,6 @@ base_model = "anthropic/claude-opus-4-1-20250805" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] +status = "deprecated" [cost] input = 15 diff --git a/providers/google-vertex-anthropic/models/claude-opus-4-5@20251101.toml b/providers/google-vertex-anthropic/models/claude-opus-4-5@20251101.toml index ba61d35c3..66b812a8c 100644 --- a/providers/google-vertex-anthropic/models/claude-opus-4-5@20251101.toml +++ b/providers/google-vertex-anthropic/models/claude-opus-4-5@20251101.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-5-20251101" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }, { type = "budget_tokens", min = 1_024 }] [cost] input = 5 diff --git a/providers/google-vertex-anthropic/models/claude-opus-4-6@default.toml b/providers/google-vertex-anthropic/models/claude-opus-4-6@default.toml index 688a5945e..f16234b69 100644 --- a/providers/google-vertex-anthropic/models/claude-opus-4-6@default.toml +++ b/providers/google-vertex-anthropic/models/claude-opus-4-6@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-6" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "max"] }, { type = "budget_tokens", min = 1_024 }] [cost] input = 5 diff --git a/providers/google-vertex-anthropic/models/claude-opus-4-7@default.toml b/providers/google-vertex-anthropic/models/claude-opus-4-7@default.toml index 65a44ac6b..b5e295391 100644 --- a/providers/google-vertex-anthropic/models/claude-opus-4-7@default.toml +++ b/providers/google-vertex-anthropic/models/claude-opus-4-7@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-7" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "xhigh", "max"] }] [cost] input = 5 diff --git a/providers/google-vertex-anthropic/models/claude-opus-4-8@default.toml b/providers/google-vertex-anthropic/models/claude-opus-4-8@default.toml index f7fe05c61..971a2dfd1 100644 --- a/providers/google-vertex-anthropic/models/claude-opus-4-8@default.toml +++ b/providers/google-vertex-anthropic/models/claude-opus-4-8@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-8" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "xhigh", "max"] }] [cost] input = 5 diff --git a/providers/google-vertex-anthropic/models/claude-opus-4@20250514.toml b/providers/google-vertex-anthropic/models/claude-opus-4@20250514.toml index 7e4fa7dd1..439593462 100644 --- a/providers/google-vertex-anthropic/models/claude-opus-4@20250514.toml +++ b/providers/google-vertex-anthropic/models/claude-opus-4@20250514.toml @@ -1,4 +1,6 @@ base_model = "anthropic/claude-opus-4-20250514" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] +status = "deprecated" [cost] input = 15 diff --git a/providers/google-vertex-anthropic/models/claude-sonnet-4-5@20250929.toml b/providers/google-vertex-anthropic/models/claude-sonnet-4-5@20250929.toml index 41b880dce..2335c64d9 100644 --- a/providers/google-vertex-anthropic/models/claude-sonnet-4-5@20250929.toml +++ b/providers/google-vertex-anthropic/models/claude-sonnet-4-5@20250929.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-sonnet-4-5-20250929" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] [cost] input = 3 diff --git a/providers/google-vertex-anthropic/models/claude-sonnet-4-6@default.toml b/providers/google-vertex-anthropic/models/claude-sonnet-4-6@default.toml index d37ba7376..a9efc40e9 100644 --- a/providers/google-vertex-anthropic/models/claude-sonnet-4-6@default.toml +++ b/providers/google-vertex-anthropic/models/claude-sonnet-4-6@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-sonnet-4-6" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "max"] }, { type = "budget_tokens", min = 1_024 }] [cost] input = 3 diff --git a/providers/google-vertex-anthropic/models/claude-sonnet-4@20250514.toml b/providers/google-vertex-anthropic/models/claude-sonnet-4@20250514.toml index 5f77a57a3..892209047 100644 --- a/providers/google-vertex-anthropic/models/claude-sonnet-4@20250514.toml +++ b/providers/google-vertex-anthropic/models/claude-sonnet-4@20250514.toml @@ -1,4 +1,6 @@ base_model = "anthropic/claude-sonnet-4-20250514" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] +status = "deprecated" [cost] input = 3 diff --git a/providers/google-vertex/models/claude-3-7-sonnet@20250219.toml b/providers/google-vertex/models/claude-3-7-sonnet@20250219.toml deleted file mode 100644 index 5cf9e8b73..000000000 --- a/providers/google-vertex/models/claude-3-7-sonnet@20250219.toml +++ /dev/null @@ -1,10 +0,0 @@ -base_model = "anthropic/claude-3-7-sonnet-20250219" - -[cost] -input = 3 -output = 15 -cache_read = 0.3 -cache_write = 3.75 - -[provider] -npm = "@ai-sdk/google-vertex/anthropic" diff --git a/providers/google-vertex/models/claude-haiku-4-5@20251001.toml b/providers/google-vertex/models/claude-haiku-4-5@20251001.toml index fbe259140..b01e7db90 100644 --- a/providers/google-vertex/models/claude-haiku-4-5@20251001.toml +++ b/providers/google-vertex/models/claude-haiku-4-5@20251001.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-haiku-4-5-20251001" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] [cost] input = 1 diff --git a/providers/google-vertex/models/claude-opus-4-1@20250805.toml b/providers/google-vertex/models/claude-opus-4-1@20250805.toml index 828785c48..d6910d233 100644 --- a/providers/google-vertex/models/claude-opus-4-1@20250805.toml +++ b/providers/google-vertex/models/claude-opus-4-1@20250805.toml @@ -1,4 +1,6 @@ base_model = "anthropic/claude-opus-4-1-20250805" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] +status = "deprecated" [cost] input = 15 diff --git a/providers/google-vertex/models/claude-opus-4-5@20251101.toml b/providers/google-vertex/models/claude-opus-4-5@20251101.toml index dde117514..e731aacd0 100644 --- a/providers/google-vertex/models/claude-opus-4-5@20251101.toml +++ b/providers/google-vertex/models/claude-opus-4-5@20251101.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-5-20251101" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }, { type = "budget_tokens", min = 1_024 }] [cost] input = 5 diff --git a/providers/google-vertex/models/claude-opus-4-6@default.toml b/providers/google-vertex/models/claude-opus-4-6@default.toml index d258b89b6..91d85ba4e 100644 --- a/providers/google-vertex/models/claude-opus-4-6@default.toml +++ b/providers/google-vertex/models/claude-opus-4-6@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-6" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "max"] }, { type = "budget_tokens", min = 1_024 }] [cost] input = 5 diff --git a/providers/google-vertex/models/claude-opus-4-7@default.toml b/providers/google-vertex/models/claude-opus-4-7@default.toml index 239ff5a53..eac0d61d5 100644 --- a/providers/google-vertex/models/claude-opus-4-7@default.toml +++ b/providers/google-vertex/models/claude-opus-4-7@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-7" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "xhigh", "max"] }] [cost] input = 5 diff --git a/providers/google-vertex/models/claude-opus-4-8@default.toml b/providers/google-vertex/models/claude-opus-4-8@default.toml index 7a76c8ee9..4d874b0fd 100644 --- a/providers/google-vertex/models/claude-opus-4-8@default.toml +++ b/providers/google-vertex/models/claude-opus-4-8@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-opus-4-8" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "xhigh", "max"] }] [cost] input = 5 diff --git a/providers/google-vertex/models/claude-opus-4@20250514.toml b/providers/google-vertex/models/claude-opus-4@20250514.toml index 426493bd0..775291855 100644 --- a/providers/google-vertex/models/claude-opus-4@20250514.toml +++ b/providers/google-vertex/models/claude-opus-4@20250514.toml @@ -1,4 +1,6 @@ base_model = "anthropic/claude-opus-4-20250514" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] +status = "deprecated" [cost] input = 15 diff --git a/providers/google-vertex/models/claude-sonnet-4-5@20250929.toml b/providers/google-vertex/models/claude-sonnet-4-5@20250929.toml index 94eb41e2b..fe76224bc 100644 --- a/providers/google-vertex/models/claude-sonnet-4-5@20250929.toml +++ b/providers/google-vertex/models/claude-sonnet-4-5@20250929.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-sonnet-4-5-20250929" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] [cost] input = 3 diff --git a/providers/google-vertex/models/claude-sonnet-4-6@default.toml b/providers/google-vertex/models/claude-sonnet-4-6@default.toml index cc1180917..14876ca38 100644 --- a/providers/google-vertex/models/claude-sonnet-4-6@default.toml +++ b/providers/google-vertex/models/claude-sonnet-4-6@default.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-sonnet-4-6" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "max"] }, { type = "budget_tokens", min = 1_024 }] [cost] input = 3 diff --git a/providers/google-vertex/models/claude-sonnet-4@20250514.toml b/providers/google-vertex/models/claude-sonnet-4@20250514.toml index e0bb6955f..00de4431f 100644 --- a/providers/google-vertex/models/claude-sonnet-4@20250514.toml +++ b/providers/google-vertex/models/claude-sonnet-4@20250514.toml @@ -1,4 +1,6 @@ base_model = "anthropic/claude-sonnet-4-20250514" +reasoning_options = [{ type = "budget_tokens", min = 1_024 }] +status = "deprecated" [cost] input = 3 diff --git a/providers/google-vertex/models/gemini-2.5-flash-lite-preview-06-17.toml b/providers/google-vertex/models/gemini-2.5-flash-lite-preview-06-17.toml deleted file mode 100644 index 6c11af151..000000000 --- a/providers/google-vertex/models/gemini-2.5-flash-lite-preview-06-17.toml +++ /dev/null @@ -1,23 +0,0 @@ -name = "Gemini 2.5 Flash Lite Preview 06-17" -family = "gemini-flash-lite" -release_date = "2025-06-17" -last_updated = "2025-06-17" -attachment = true -reasoning = true -temperature = true -knowledge = "2025-01" -tool_call = true -open_weights = false - -[cost] -input = 0.10 -output = 0.40 -cache_read = 0.025 - -[limit] -context = 65_536 -output = 65_536 - -[modalities] -input = ["text", "image", "audio", "video", "pdf"] -output = ["text"] diff --git a/providers/google-vertex/models/gemini-2.5-flash-lite.toml b/providers/google-vertex/models/gemini-2.5-flash-lite.toml index ffa9f61e7..ed8487476 100644 --- a/providers/google-vertex/models/gemini-2.5-flash-lite.toml +++ b/providers/google-vertex/models/gemini-2.5-flash-lite.toml @@ -1,5 +1,6 @@ base_model = "google/gemini-2.5-flash-lite" base_model_omit = ["structured_output"] +reasoning_options = [{ type = "toggle" }, { type = "budget_tokens", min = 512, max = 24_576 }] [cost] input = 0.1 diff --git a/providers/google-vertex/models/gemini-2.5-flash-preview-09-2025.toml b/providers/google-vertex/models/gemini-2.5-flash-preview-09-2025.toml deleted file mode 100644 index 0f71474c9..000000000 --- a/providers/google-vertex/models/gemini-2.5-flash-preview-09-2025.toml +++ /dev/null @@ -1,24 +0,0 @@ -name = "Gemini 2.5 Flash Preview 09-25" -family = "gemini-flash" -release_date = "2025-09-25" -last_updated = "2025-09-25" -attachment = true -reasoning = true -temperature = true -knowledge = "2025-01" -tool_call = true -open_weights = false - -[cost] -input = 0.30 -output = 2.50 -cache_read = 0.075 -cache_write = 0.383 - -[limit] -context = 1_048_576 -output = 65_536 - -[modalities] -input = ["text", "image", "audio", "video", "pdf"] -output = ["text"] diff --git a/providers/google-vertex/models/gemini-2.5-flash.toml b/providers/google-vertex/models/gemini-2.5-flash.toml index c592c6bd3..df0ff49f6 100644 --- a/providers/google-vertex/models/gemini-2.5-flash.toml +++ b/providers/google-vertex/models/gemini-2.5-flash.toml @@ -4,6 +4,7 @@ release_date = "2025-06-17" last_updated = "2025-06-17" attachment = true reasoning = true +reasoning_options = [{ type = "toggle" }, { type = "budget_tokens", min = 0, max = 24_576 }] temperature = true knowledge = "2025-01" tool_call = true diff --git a/providers/google-vertex/models/gemini-2.5-pro.toml b/providers/google-vertex/models/gemini-2.5-pro.toml index 8e228b72e..84bf70c6e 100644 --- a/providers/google-vertex/models/gemini-2.5-pro.toml +++ b/providers/google-vertex/models/gemini-2.5-pro.toml @@ -1,5 +1,6 @@ base_model = "google/gemini-2.5-pro" base_model_omit = ["structured_output"] +reasoning_options = [{ type = "budget_tokens", min = 128, max = 32_768 }] [cost] input = 1.25 diff --git a/providers/google-vertex/models/gemini-3-flash-preview.toml b/providers/google-vertex/models/gemini-3-flash-preview.toml index 83ddd3dd5..56a17f392 100644 --- a/providers/google-vertex/models/gemini-3-flash-preview.toml +++ b/providers/google-vertex/models/gemini-3-flash-preview.toml @@ -1,4 +1,5 @@ base_model = "google/gemini-3-flash-preview" +reasoning_options = [{ type = "effort", values = ["minimal", "low", "medium", "high"] }] [cost] input = 0.5 diff --git a/providers/google-vertex/models/gemini-3-pro-preview.toml b/providers/google-vertex/models/gemini-3-pro-preview.toml deleted file mode 100644 index 422582e33..000000000 --- a/providers/google-vertex/models/gemini-3-pro-preview.toml +++ /dev/null @@ -1,30 +0,0 @@ -name = "Gemini 3 Pro Preview" -family = "gemini-pro" -release_date = "2025-11-18" -last_updated = "2025-11-18" -attachment = true -reasoning = true -temperature = true -tool_call = true -structured_output = true -knowledge = "2025-01" -open_weights = false - -[cost] -input = 2.00 -output = 12.00 -cache_read = 0.20 - -[[cost.tiers]] -tier = { size = 200_000 } -input = 4.00 -output = 18.00 -cache_read = 0.40 - -[limit] -context = 1_048_576 -output = 65_536 - -[modalities] -input = ["text", "image", "video", "audio", "pdf"] -output = ["text"] diff --git a/providers/google-vertex/models/gemini-3.1-flash-lite-preview.toml b/providers/google-vertex/models/gemini-3.1-flash-lite-preview.toml index 65e9c4c1d..e862d7dcb 100644 --- a/providers/google-vertex/models/gemini-3.1-flash-lite-preview.toml +++ b/providers/google-vertex/models/gemini-3.1-flash-lite-preview.toml @@ -1,4 +1,5 @@ base_model = "google/gemini-3.1-flash-lite-preview" +reasoning_options = [{ type = "effort", values = ["minimal", "low", "medium", "high"] }] [cost] input = 0.25 diff --git a/providers/google-vertex/models/gemini-3.1-flash-lite.toml b/providers/google-vertex/models/gemini-3.1-flash-lite.toml index 419b0ef87..d9d9b5d7d 100644 --- a/providers/google-vertex/models/gemini-3.1-flash-lite.toml +++ b/providers/google-vertex/models/gemini-3.1-flash-lite.toml @@ -1,4 +1,5 @@ base_model = "google/gemini-3.1-flash-lite" +reasoning_options = [{ type = "effort", values = ["minimal", "low", "medium", "high"] }] [cost] input = 0.25 diff --git a/providers/google-vertex/models/gemini-3.1-pro-preview-customtools.toml b/providers/google-vertex/models/gemini-3.1-pro-preview-customtools.toml index ac88bf0cb..6669e542f 100644 --- a/providers/google-vertex/models/gemini-3.1-pro-preview-customtools.toml +++ b/providers/google-vertex/models/gemini-3.1-pro-preview-customtools.toml @@ -1,4 +1,5 @@ base_model = "google/gemini-3.1-pro-preview-customtools" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }] [cost] input = 2 diff --git a/providers/google-vertex/models/gemini-3.1-pro-preview.toml b/providers/google-vertex/models/gemini-3.1-pro-preview.toml index 30cc13476..e2b3bf54f 100644 --- a/providers/google-vertex/models/gemini-3.1-pro-preview.toml +++ b/providers/google-vertex/models/gemini-3.1-pro-preview.toml @@ -1,4 +1,5 @@ base_model = "google/gemini-3.1-pro-preview" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }] [cost] input = 2 diff --git a/providers/google-vertex/models/gemini-3.5-flash.toml b/providers/google-vertex/models/gemini-3.5-flash.toml index 78283f7dd..0d75d4a0d 100644 --- a/providers/google-vertex/models/gemini-3.5-flash.toml +++ b/providers/google-vertex/models/gemini-3.5-flash.toml @@ -1,4 +1,5 @@ base_model = "google/gemini-3.5-flash" +reasoning_options = [{ type = "effort", values = ["minimal", "low", "medium", "high"] }] [cost] input = 1.5 diff --git a/providers/google-vertex/models/gemini-flash-latest.toml b/providers/google-vertex/models/gemini-flash-latest.toml deleted file mode 100644 index c4a3785c1..000000000 --- a/providers/google-vertex/models/gemini-flash-latest.toml +++ /dev/null @@ -1,24 +0,0 @@ -name = "Gemini Flash Latest" -family = "gemini-flash" -release_date = "2025-09-25" -last_updated = "2025-09-25" -attachment = true -reasoning = true -temperature = true -knowledge = "2025-01" -tool_call = true -open_weights = false - -[cost] -input = 0.30 -output = 2.50 -cache_read = 0.075 -cache_write = 0.383 - -[limit] -context = 1_048_576 -output = 65_536 - -[modalities] -input = ["text", "image", "audio", "video", "pdf"] -output = ["text"] diff --git a/providers/google-vertex/models/gemini-flash-lite-latest.toml b/providers/google-vertex/models/gemini-flash-lite-latest.toml deleted file mode 100644 index 3d8f886fe..000000000 --- a/providers/google-vertex/models/gemini-flash-lite-latest.toml +++ /dev/null @@ -1,7 +0,0 @@ -base_model = "google/gemini-flash-lite-latest" -base_model_omit = ["structured_output"] - -[cost] -input = 0.1 -output = 0.4 -cache_read = 0.025 From ffedd884f7ab0f67edecf541945dac31afb09464 Mon Sep 17 00:00:00 2001 From: Aiden Cline Date: Mon, 8 Jun 2026 22:06:17 -0500 Subject: [PATCH 2/3] fix(google-vertex): remove retired models --- .../models/claude-3-5-haiku@20241022.toml | 1 + .../models/claude-3-5-sonnet@20241022.toml | 7 ------ .../models/claude-3-5-haiku@20241022.toml | 1 + .../models/claude-3-5-sonnet@20241022.toml | 10 -------- .../models/gemini-2.0-flash-lite.toml | 6 ----- .../models/gemini-2.0-flash.toml | 23 ------------------- 6 files changed, 2 insertions(+), 46 deletions(-) delete mode 100644 providers/google-vertex-anthropic/models/claude-3-5-sonnet@20241022.toml delete mode 100644 providers/google-vertex/models/claude-3-5-sonnet@20241022.toml delete mode 100644 providers/google-vertex/models/gemini-2.0-flash-lite.toml delete mode 100644 providers/google-vertex/models/gemini-2.0-flash.toml diff --git a/providers/google-vertex-anthropic/models/claude-3-5-haiku@20241022.toml b/providers/google-vertex-anthropic/models/claude-3-5-haiku@20241022.toml index 187db9328..2c14ab829 100644 --- a/providers/google-vertex-anthropic/models/claude-3-5-haiku@20241022.toml +++ b/providers/google-vertex-anthropic/models/claude-3-5-haiku@20241022.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-3-5-haiku-20241022" +status = "deprecated" [cost] input = 0.8 diff --git a/providers/google-vertex-anthropic/models/claude-3-5-sonnet@20241022.toml b/providers/google-vertex-anthropic/models/claude-3-5-sonnet@20241022.toml deleted file mode 100644 index 1c9c5bc08..000000000 --- a/providers/google-vertex-anthropic/models/claude-3-5-sonnet@20241022.toml +++ /dev/null @@ -1,7 +0,0 @@ -base_model = "anthropic/claude-3-5-sonnet-20241022" - -[cost] -input = 3 -output = 15 -cache_read = 0.3 -cache_write = 3.75 diff --git a/providers/google-vertex/models/claude-3-5-haiku@20241022.toml b/providers/google-vertex/models/claude-3-5-haiku@20241022.toml index 0ca47a7c2..ba2cc2dae 100644 --- a/providers/google-vertex/models/claude-3-5-haiku@20241022.toml +++ b/providers/google-vertex/models/claude-3-5-haiku@20241022.toml @@ -1,4 +1,5 @@ base_model = "anthropic/claude-3-5-haiku-20241022" +status = "deprecated" [cost] input = 0.8 diff --git a/providers/google-vertex/models/claude-3-5-sonnet@20241022.toml b/providers/google-vertex/models/claude-3-5-sonnet@20241022.toml deleted file mode 100644 index 47e5f3450..000000000 --- a/providers/google-vertex/models/claude-3-5-sonnet@20241022.toml +++ /dev/null @@ -1,10 +0,0 @@ -base_model = "anthropic/claude-3-5-sonnet-20241022" - -[cost] -input = 3 -output = 15 -cache_read = 0.3 -cache_write = 3.75 - -[provider] -npm = "@ai-sdk/google-vertex/anthropic" diff --git a/providers/google-vertex/models/gemini-2.0-flash-lite.toml b/providers/google-vertex/models/gemini-2.0-flash-lite.toml deleted file mode 100644 index f8e1843fa..000000000 --- a/providers/google-vertex/models/gemini-2.0-flash-lite.toml +++ /dev/null @@ -1,6 +0,0 @@ -base_model = "google/gemini-2.0-flash-lite" -base_model_omit = ["structured_output"] - -[cost] -input = 0.075 -output = 0.3 diff --git a/providers/google-vertex/models/gemini-2.0-flash.toml b/providers/google-vertex/models/gemini-2.0-flash.toml deleted file mode 100644 index 3f7a2bb0a..000000000 --- a/providers/google-vertex/models/gemini-2.0-flash.toml +++ /dev/null @@ -1,23 +0,0 @@ -name = "Gemini 2.0 Flash" -family = "gemini-flash" -release_date = "2024-12-11" -last_updated = "2024-12-11" -attachment = true -reasoning = false -temperature = true -knowledge = "2024-06" -tool_call = true -open_weights = false - -[cost] -input = 0.15 -output = 0.60 -cache_read = 0.025 - -[limit] -context = 1_048_576 -output = 8_192 - -[modalities] -input = ["text", "image", "audio", "video", "pdf"] -output = ["text"] From dcae17ea266d4b686dccd3ffaff2846be3ce592f Mon Sep 17 00:00:00 2001 From: Aiden Cline Date: Mon, 8 Jun 2026 22:12:00 -0500 Subject: [PATCH 3/3] fix(google-vertex): retain latest Gemini aliases --- .../models/gemini-flash-latest.toml | 25 +++++++++++++++++++ .../models/gemini-flash-lite-latest.toml | 8 ++++++ 2 files changed, 33 insertions(+) create mode 100644 providers/google-vertex/models/gemini-flash-latest.toml create mode 100644 providers/google-vertex/models/gemini-flash-lite-latest.toml diff --git a/providers/google-vertex/models/gemini-flash-latest.toml b/providers/google-vertex/models/gemini-flash-latest.toml new file mode 100644 index 000000000..9fe2fb3b4 --- /dev/null +++ b/providers/google-vertex/models/gemini-flash-latest.toml @@ -0,0 +1,25 @@ +name = "Gemini Flash Latest" +family = "gemini-flash" +release_date = "2025-09-25" +last_updated = "2025-09-25" +attachment = true +reasoning = true +reasoning_options = [{ type = "effort", values = ["minimal", "low", "medium", "high"] }] +temperature = true +knowledge = "2025-01" +tool_call = true +open_weights = false + +[cost] +input = 0.30 +output = 2.50 +cache_read = 0.075 +cache_write = 0.383 + +[limit] +context = 1_048_576 +output = 65_536 + +[modalities] +input = ["text", "image", "audio", "video", "pdf"] +output = ["text"] diff --git a/providers/google-vertex/models/gemini-flash-lite-latest.toml b/providers/google-vertex/models/gemini-flash-lite-latest.toml new file mode 100644 index 000000000..3c5645b4c --- /dev/null +++ b/providers/google-vertex/models/gemini-flash-lite-latest.toml @@ -0,0 +1,8 @@ +base_model = "google/gemini-flash-lite-latest" +base_model_omit = ["structured_output"] +reasoning_options = [{ type = "effort", values = ["minimal", "low", "medium", "high"] }] + +[cost] +input = 0.1 +output = 0.4 +cache_read = 0.025