Feature/models support expansion

CHANGELOG.md

@@ -8,6 +8,43 @@ and the project adheres to [Semantic Versioning 2.0.0](https://semver.org/spec/v
 > Tags prior to **v0.4.0** were cut in the private repository and produced no
 > public artifacts; the first publicly released version is v0.4.0.
 
+## [1.4.1] - 2026-05-30
+
+### Changed
+- **`setup` no longer forces an API-key re-entry when one already exists.**
+  Re-running `commitbrief setup` for a provider that already has a key in the
+  target config now lets you leave the key prompt blank to keep the stored
+  key, so switching only the active provider/model doesn't require retyping
+  credentials. First-time configuration (no existing key) still requires a
+  non-empty key. For a non-interactive path, `providers use <name>` switches
+  the active provider and `config set providers.<name>.model <model>` changes
+  the model — both leave API keys untouched.
+
+### Added
+- **OpenAI: GPT-5 family support — `gpt-5.5`, `gpt-5.4-mini`, `gpt-5.5-pro`.**
+  All three are selectable in `commitbrief setup` and carry correct context
+  windows (1.05M / 400K / 1.05M input) and pricing. The new default OpenAI
+  model is **`gpt-5.4-mini`** (was `gpt-4o`); existing configs that pin a
+  model are unaffected. `gpt-5.5-pro` is served through OpenAI's Responses
+  API (it is not available on Chat Completions) and may take several minutes
+  per review. Reasoning models get a larger default output-token budget so a
+  findings JSON isn't truncated by reasoning tokens.
+- **Gemini: `gemini-3.5-flash` support.**
+
+### Changed
+- **Gemini model lineup refreshed to the 3.x family.** `gemini-2.5-pro` →
+  `gemini-3.1-pro-preview`, `gemini-2.5-flash` → `gemini-3.5-flash`,
+  `gemini-1.5-flash` → `gemini-3.1-flash-lite`. The new default Gemini model
+  is **`gemini-3.5-flash`** (was `gemini-2.5-pro`). Pricing and context
+  windows updated; `gemini-3.1-pro-preview` carries tiered pricing (the
+  ≤200K-token base tier is snapshotted). Configs pinning a removed 2.x/1.5
+  model should switch to a 3.x ID.
+- **Anthropic: `claude-opus-4-7` → `claude-opus-4-8`.** Opus 4.8 is now the
+  default Anthropic model. Its built-in pricing ($5 input / $25 output / $0.50
+  cache-read per 1M) and context window (1M tokens, up from the 200K the old
+  table recorded for 4.7) are updated to match Anthropic's current model docs.
+  Configs pinning `claude-opus-4-7` should switch to `claude-opus-4-8`.
+
 ## [1.4.0] - 2026-05-29
 
 ### Fixed

README.md

@@ -260,9 +260,9 @@ Four API providers + two CLI-tool-backed providers ship in the box:
 
 | Provider | Models | Notes |
 |----------|--------|-------|
-| **Anthropic** | Claude Opus 4.7, Sonnet 4.6, Haiku 4.5 | Ephemeral prompt caching (5 m TTL) cuts repeated input cost ~10×. |
-| **OpenAI** | GPT-4o, GPT-4o-mini | Automatic prompt caching at ≥1024-token prefixes. |
-| **Google Gemini** | Gemini 2.5 Pro (2 M context!), 2.5 Flash, 1.5 Flash | Largest free-tier context windows. |
+| **Anthropic** | Claude Opus 4.8 (default), Sonnet 4.6, Haiku 4.5 | Ephemeral prompt caching (5 m TTL) cuts repeated input cost ~10×. Opus 4.8 advertises a 1 M-token context. |
+| **OpenAI** | GPT-5.4-mini (default), GPT-5.5, GPT-5.5-pro, GPT-4o, GPT-4o-mini | Automatic prompt caching at ≥1024-token prefixes. `gpt-5.5-pro` runs via the Responses API (not Chat Completions) and can take several minutes per review. |
+| **Google Gemini** | Gemini 3.5 Flash (default), 3.1 Pro, 3.1 Flash-Lite | ~1 M-token context windows. `gemini-3.1-pro-preview` is a preview model. |
 | **DeepSeek** | deepseek-chat, deepseek-reasoner | OpenAI-compatible API (`DEEPSEEK_API_KEY`); JSON is prompt-driven (degrades gracefully). |
 | **Mistral** | Mistral Large / Small, Codestral | OpenAI-compatible API (`MISTRAL_API_KEY`). |
 | **Cohere** | Command R+ / R, Command A | Cohere's OpenAI-compatibility endpoint (`COHERE_API_KEY`). |
@@ -361,7 +361,7 @@ Two-tier YAML config with field-level merge:
 
 Plus environment variables for credentials and runtime tweaks, and
 CLI flags for one-off overrides (`--provider gemini --model
-gemini-2.5-flash`).
+gemini-3.5-flash`).
 
 | Variable | Effect |
 |---|---|
@@ -379,13 +379,13 @@ version: 1
 provider: anthropic                # default provider
 providers:
   anthropic:
-    model: claude-opus-4-7
+    model: claude-opus-4-8
     pricing:                         # optional: override built-in $/1M rates
-      claude-opus-4-7:               # (cost preflight / verbose footer / cache)
-        input_per_1m: 15.0
-        output_per_1m: 75.0          # omitted fields keep the built-in value
+      claude-opus-4-8:               # (cost preflight / verbose footer / cache)
+        input_per_1m: 5.0
+        output_per_1m: 25.0          # omitted fields keep the built-in value
   openai:
-    model: gpt-4o
+    model: gpt-5.4-mini
   ollama:
     model: qwen2.5-coder:14b
     base_url: http://localhost:11434

internal/config/defaults.go

@@ -7,9 +7,9 @@ func Default() *Config {
 		Version:  CurrentSchemaVersion,
 		Provider: "anthropic",
 		Providers: map[string]ProviderConfig{
-			"anthropic": {Model: "claude-opus-4-7", BaseURL: "https://api.anthropic.com"},
-			"openai":    {Model: "gpt-4o", BaseURL: "https://api.openai.com/v1"},
-			"gemini":    {Model: "gemini-2.5-pro"},
+			"anthropic": {Model: "claude-opus-4-8", BaseURL: "https://api.anthropic.com"},
+			"openai":    {Model: "gpt-5.4-mini", BaseURL: "https://api.openai.com/v1"},
+			"gemini":    {Model: "gemini-3.5-flash"},
 			"ollama":    {Model: "qwen2.5-coder:14b", BaseURL: "http://localhost:11434"},
 		},
 		Output: OutputConfig{

internal/i18n/messages.en.yml

@@ -3,6 +3,8 @@ setup.provider.prompt: "Which provider would you like to configure?"
 setup.provider.help: "Pick the LLM provider you have an API key for."
 setup.api_key.prompt: "Enter your API key:"
 setup.api_key.empty: "API key cannot be empty."
+setup.api_key.prompt_keep: "Enter a new API key (leave blank to keep the existing one):"
+setup.api_key.help_keep: "A key is already configured for this provider. Leave blank to keep it, or enter a new one to replace it."
 setup.base_url.prompt: "Ollama base URL:"
 setup.model.prompt: "Pick a model:"
 setup.model.discover_failed: "Model name (could not discover from Ollama):"

internal/i18n/messages.tr.yml

@@ -3,6 +3,8 @@ setup.provider.prompt: "Hangi sağlayıcıyı yapılandırmak istersiniz?"
 setup.provider.help: "API anahtarınız olan LLM sağlayıcısını seçin."
 setup.api_key.prompt: "API anahtarınızı girin:"
 setup.api_key.empty: "API anahtarı boş olamaz."
+setup.api_key.prompt_keep: "Yeni bir API anahtarı girin (mevcut anahtarı korumak için boş bırakın):"
+setup.api_key.help_keep: "Bu sağlayıcı için zaten bir anahtar yapılandırılmış. Korumak için boş bırakın ya da değiştirmek için yenisini girin."
 setup.base_url.prompt: "Ollama temel URL'si:"
 setup.model.prompt: "Bir model seçin:"
 setup.model.discover_failed: "Model adı (Ollama'dan keşfedilemedi):"

internal/provider/anthropic/anthropic_test.go

@@ -21,7 +21,7 @@ func TestModelsList(t *testing.T) {
 		t.Errorf("Models() length = %d, want 3", len(got))
 	}
 	want := map[string]bool{
-		ModelOpus47:   true,
+		ModelOpus48:   true,
 		ModelSonnet46: true,
 		ModelHaiku45:  true,
 	}
@@ -41,8 +41,8 @@ func TestModelsDefensiveCopy(t *testing.T) {
 }
 
 func TestIsModelSupported(t *testing.T) {
-	if !IsModelSupported(ModelOpus47) {
-		t.Error("Opus 4.7 should be supported")
+	if !IsModelSupported(ModelOpus48) {
+		t.Error("Opus 4.8 should be supported")
 	}
 	if IsModelSupported("gpt-4o") {
 		t.Error("OpenAI model should not be reported as supported")
@@ -53,8 +53,8 @@ func TestIsModelSupported(t *testing.T) {
 }
 
 func TestContextWindow(t *testing.T) {
-	if contextWindowFor(ModelOpus47) != 200_000 {
-		t.Error("Opus 4.7 context window wrong")
+	if contextWindowFor(ModelOpus48) != 1_000_000 {
+		t.Error("Opus 4.8 should advertise the 1M window")
 	}
 	if contextWindowFor(ModelSonnet46) != 1_000_000 {
 		t.Error("Sonnet 4.6 should advertise the 1M window")
@@ -65,7 +65,7 @@ func TestContextWindow(t *testing.T) {
 }
 
 func TestPricingLookup(t *testing.T) {
-	p := pricingFor(ModelOpus47)
+	p := pricingFor(ModelOpus48)
 	if p.InputPer1M == 0 || p.OutputPer1M == 0 {
 		t.Errorf("Opus pricing missing: %+v", p)
 	}
@@ -223,7 +223,7 @@ func messagePayload(content []map[string]any, inputTokens, outputTokens int) map
 		"id":            "msg_test",
 		"type":          "message",
 		"role":          "assistant",
-		"model":         ModelOpus47,
+		"model":         ModelOpus48,
 		"content":       content,
 		"stop_reason":   "end_turn",
 		"stop_sequence": nil,
@@ -253,7 +253,7 @@ func TestReviewWithFakeServerDegradesToText(t *testing.T) {
 		t.Fatal(err)
 	}
 	resp, err := c.Review(context.Background(), provider.Request{
-		Model:        ModelOpus47,
+		Model:        ModelOpus48,
 		SystemPrompt: "rules",
 		UserPrompt:   "diff",
 		MaxTokens:    256,
@@ -291,7 +291,7 @@ func TestReviewWithToolUseFakeServer(t *testing.T) {
 		t.Fatal(err)
 	}
 	resp, err := c.Review(context.Background(), provider.Request{
-		Model:        ModelOpus47,
+		Model:        ModelOpus48,
 		SystemPrompt: "rules",
 		UserPrompt:   "diff",
 		MaxTokens:    256,

internal/provider/anthropic/context_window.go

@@ -5,7 +5,7 @@ package anthropic
 const defaultContextWindow = 200_000
 
 var contextWindows = map[string]int{
-	ModelOpus47:   200_000,
+	ModelOpus48:   1_000_000,
 	ModelSonnet46: 1_000_000,
 	ModelHaiku45:  200_000,
 }

internal/provider/anthropic/models.go

@@ -5,15 +5,15 @@ package anthropic
 const (
 	Name = "anthropic"
 
-	ModelOpus47   = "claude-opus-4-7"
+	ModelOpus48   = "claude-opus-4-8"
 	ModelSonnet46 = "claude-sonnet-4-6"
 	ModelHaiku45  = "claude-haiku-4-5-20251001"
 
-	DefaultModel = ModelOpus47
+	DefaultModel = ModelOpus48
 )
 
 var supportedModels = []string{
-	ModelOpus47,
+	ModelOpus48,
 	ModelSonnet46,
 	ModelHaiku45,
 }

internal/provider/anthropic/pricing.go

@@ -9,10 +9,10 @@ import "github.com/CommitBrief/commitbrief/internal/provider"
 // should be refreshed when a model price changes. Cached input is the
 // "prompt cache read" rate per Anthropic's ephemeral cache discount.
 var pricingTable = map[string]provider.Pricing{
-	ModelOpus47: {
-		InputPer1M:       15.00,
-		OutputPer1M:      75.00,
-		CachedInputPer1M: 1.50,
+	ModelOpus48: {
+		InputPer1M:       5.00,
+		OutputPer1M:      25.00,
+		CachedInputPer1M: 0.50,
 	},
 	ModelSonnet46: {
 		InputPer1M:       3.00,

internal/provider/gemini/context_window.go

@@ -4,10 +4,14 @@ package gemini
 
 const defaultContextWindow = 1_000_000
 
+// Gemini 3.x family baseline input window is 1M tokens. Exact published
+// limits for these preview models were not yet broken out in the model
+// docs at integration time; 1M is the documented floor and the safe value
+// for the over-context guard. Bump if Google publishes larger windows.
 var contextWindows = map[string]int{
-	ModelPro2_5:   2_000_000,
-	ModelFlash2_5: 1_000_000,
-	ModelFlash1_5: 1_000_000,
+	ModelPro31:       1_000_000,
+	ModelFlash35:     1_000_000,
+	ModelFlashLite31: 1_000_000,
 }
 
 func contextWindowFor(model string) int {

internal/provider/gemini/gemini_test.go

@@ -32,25 +32,25 @@ func TestModelsDefensiveCopy(t *testing.T) {
 }
 
 func TestIsModelSupported(t *testing.T) {
-	if !IsModelSupported(ModelPro2_5) {
-		t.Error("gemini-2.5-pro should be supported")
+	if !IsModelSupported(ModelPro31) {
+		t.Error("gemini-3.1-pro-preview should be supported")
 	}
 	if IsModelSupported("gpt-4o") {
 		t.Error("OpenAI model should not be supported here")
 	}
 }
 
 func TestContextWindow(t *testing.T) {
-	if contextWindowFor(ModelPro2_5) != 2_000_000 {
-		t.Errorf("pro 2.5 context window wrong: %d", contextWindowFor(ModelPro2_5))
+	if contextWindowFor(ModelPro31) != 1_000_000 {
+		t.Errorf("pro 3.1 context window wrong: %d", contextWindowFor(ModelPro31))
 	}
 	if contextWindowFor("unknown") != defaultContextWindow {
 		t.Error("unknown model should fall back to defaultContextWindow")
 	}
 }
 
 func TestPricingLookup(t *testing.T) {
-	p := pricingFor(ModelPro2_5)
+	p := pricingFor(ModelPro31)
 	if p.InputPer1M == 0 || p.OutputPer1M == 0 {
 		t.Errorf("pro pricing missing: %+v", p)
 	}
@@ -138,7 +138,7 @@ func TestReviewWithFakeServer(t *testing.T) {
 		t.Fatal(err)
 	}
 	resp, err := c.Review(context.Background(), provider.Request{
-		Model:        ModelPro2_5,
+		Model:        ModelPro31,
 		SystemPrompt: "rules",
 		UserPrompt:   "diff",
 		MaxTokens:    256,

internal/provider/gemini/models.go

@@ -5,17 +5,20 @@ package gemini
 const (
 	Name = "gemini"
 
-	ModelPro2_5   = "gemini-2.5-pro"
-	ModelFlash2_5 = "gemini-2.5-flash"
-	ModelFlash1_5 = "gemini-1.5-flash"
+	// ModelPro31 is a preview model: the public pricing page lists the Pro
+	// tier only as gemini-3.1-pro-preview. Update to the stable ID once
+	// Google promotes it out of preview.
+	ModelPro31       = "gemini-3.1-pro-preview"
+	ModelFlash35     = "gemini-3.5-flash"
+	ModelFlashLite31 = "gemini-3.1-flash-lite"
 
-	DefaultModel = ModelPro2_5
+	DefaultModel = ModelFlash35
 )
 
 var supportedModels = []string{
-	ModelPro2_5,
-	ModelFlash2_5,
-	ModelFlash1_5,
+	ModelPro31,
+	ModelFlash35,
+	ModelFlashLite31,
 }
 
 func Models() []string {

internal/provider/gemini/pricing.go

@@ -4,26 +4,28 @@ package gemini
 
 import "github.com/CommitBrief/commitbrief/internal/provider"
 
-// Gemini per-1M-token pricing snapshot (paid tier).
+// Gemini per-1M-token pricing snapshot (paid "Standard" tier).
 // Source: https://ai.google.dev/gemini-api/docs/pricing
-// CachedInputPer1M reflects Gemini's context-caching discount (separate
-// API; we track the per-token rate so cost reporting is accurate when a
-// cache is wired up in a future phase).
+// gemini-3.1-pro-preview has tiered pricing (≤200K vs >200K input tokens:
+// $2/$12 vs $4/$18); we snapshot the ≤200K base, so the verbose cost footer
+// may under-report on very large inputs. CachedInputPer1M reflects Gemini's
+// implicit-cache read discount (~0.25× input); context caching is not wired
+// into the client yet, so the rate is informational for now.
 var pricingTable = map[string]provider.Pricing{
-	ModelPro2_5: {
-		InputPer1M:       1.25,
-		OutputPer1M:      10.00,
-		CachedInputPer1M: 0.31,
+	ModelPro31: {
+		InputPer1M:       2.00,
+		OutputPer1M:      12.00,
+		CachedInputPer1M: 0.50,
 	},
-	ModelFlash2_5: {
-		InputPer1M:       0.30,
-		OutputPer1M:      2.50,
-		CachedInputPer1M: 0.075,
+	ModelFlash35: {
+		InputPer1M:       1.50,
+		OutputPer1M:      9.00,
+		CachedInputPer1M: 0.375,
 	},
-	ModelFlash1_5: {
-		InputPer1M:       0.075,
-		OutputPer1M:      0.30,
-		CachedInputPer1M: 0.01875,
+	ModelFlashLite31: {
+		InputPer1M:       0.25,
+		OutputPer1M:      1.50,
+		CachedInputPer1M: 0.0625,
 	},
 }