From 49510f5d856c6ea90a2928202b2ec09e99d91c2a Mon Sep 17 00:00:00 2001 From: Sergey Kozyrenko Date: Mon, 13 Jul 2026 02:22:08 +0700 Subject: [PATCH] chore(llms): refresh default models to current generations anthropic: claude-sonnet-4-5 -> claude-sonnet-5 (note: Sonnet 5 rejects non-default sampling params, so callers that relied on the default model while setting a custom temperature must now pick a model explicitly); openai: gpt-4.1-mini -> gpt-5.4-mini (same efficiency tier); googleai: gemini-2.0-flash -> gemini-2.5-flash (staying on the thinkingBudget-compatible generation until thinking_level support lands). Co-Authored-By: Claude Fable 5 --- llms/anthropic/internal/anthropicclient/anthropicclient.go | 2 +- llms/googleai/googleai_unit_test.go | 2 +- llms/googleai/option.go | 2 +- llms/googleai/vertex/new_test.go | 4 ++-- llms/openai/internal/openaiclient/chat.go | 2 +- 5 files changed, 6 insertions(+), 6 deletions(-) diff --git a/llms/anthropic/internal/anthropicclient/anthropicclient.go b/llms/anthropic/internal/anthropicclient/anthropicclient.go index b91fddb6..291734eb 100644 --- a/llms/anthropic/internal/anthropicclient/anthropicclient.go +++ b/llms/anthropic/internal/anthropicclient/anthropicclient.go @@ -17,7 +17,7 @@ import ( const ( DefaultBaseURL = "https://api.anthropic.com/v1" - defaultModel = "claude-sonnet-4-5" + defaultModel = "claude-sonnet-5" ) // ErrEmptyResponse is returned when the Anthropic API returns an empty response. diff --git a/llms/googleai/googleai_unit_test.go b/llms/googleai/googleai_unit_test.go index b31973a2..5b968491 100644 --- a/llms/googleai/googleai_unit_test.go +++ b/llms/googleai/googleai_unit_test.go @@ -88,7 +88,7 @@ func TestDefaultOptions(t *testing.T) { opts := DefaultOptions() - assert.Equal(t, "gemini-2.0-flash", opts.DefaultModel) + assert.Equal(t, "gemini-2.5-flash", opts.DefaultModel) assert.Equal(t, "gemini-embedding-001", opts.DefaultEmbeddingModel) assert.Equal(t, 1, opts.DefaultCandidateCount) assert.Equal(t, 2048, opts.DefaultMaxTokens) diff --git a/llms/googleai/option.go b/llms/googleai/option.go index 7e2a4cf3..b2503876 100644 --- a/llms/googleai/option.go +++ b/llms/googleai/option.go @@ -34,7 +34,7 @@ func DefaultOptions() Options { return Options{ CloudProject: "", CloudLocation: "", - DefaultModel: "gemini-2.0-flash", + DefaultModel: "gemini-2.5-flash", DefaultEmbeddingModel: "gemini-embedding-001", DefaultCandidateCount: 1, DefaultMaxTokens: 2048, diff --git a/llms/googleai/vertex/new_test.go b/llms/googleai/vertex/new_test.go index 2faf622e..d0f2b6e3 100644 --- a/llms/googleai/vertex/new_test.go +++ b/llms/googleai/vertex/new_test.go @@ -19,8 +19,8 @@ func TestDefaultOptions(t *testing.T) { opts := googleai.DefaultOptions() // Test default values - if opts.DefaultModel != "gemini-2.0-flash" { - t.Errorf("expected default model 'gemini-2.0-flash', got %q", opts.DefaultModel) + if opts.DefaultModel != "gemini-2.5-flash" { + t.Errorf("expected default model 'gemini-2.5-flash', got %q", opts.DefaultModel) } if opts.DefaultEmbeddingModel != "gemini-embedding-001" { t.Errorf("expected default embedding model 'gemini-embedding-001', got %q", opts.DefaultEmbeddingModel) diff --git a/llms/openai/internal/openaiclient/chat.go b/llms/openai/internal/openaiclient/chat.go index caa53076..a6adb45c 100644 --- a/llms/openai/internal/openaiclient/chat.go +++ b/llms/openai/internal/openaiclient/chat.go @@ -18,7 +18,7 @@ import ( ) const ( - defaultChatModel = "gpt-4.1-mini" + defaultChatModel = "gpt-5.4-mini" ) var ErrContentExclusive = errors.New("only one of Content / MultiContent allowed in message")