🤖 refactor: centralize model constants + fix previousResponseId bug

src/constants/knownModels.ts

@@ -0,0 +1,125 @@
+/**
+ * Centralized model metadata. Update model versions here and everywhere else will follow.
+ */
+
+type ModelProvider = "anthropic" | "openai";
+
+interface KnownModelDefinition {
+  /** Provider identifier used by SDK factories */
+  provider: ModelProvider;
+  /** Provider-specific model name (no provider prefix) */
+  providerModelId: string;
+  /** Aliases that should resolve to this model */
+  aliases?: string[];
+  /** Preload tokenizer encodings at startup */
+  warm?: boolean;
+  /** Use as global default model */
+  isDefault?: boolean;
+  /** Optional tokenizer override for ai-tokenizer */
+  tokenizerOverride?: string;
+}
+
+interface KnownModel extends KnownModelDefinition {
+  /** Full model id string in the format provider:model */
+  id: `${ModelProvider}:${string}`;
+}
+
+// Model definitions. Note we avoid listing legacy models here. These represent the focal models
+// of the community.
+const MODEL_DEFINITIONS = {
+  SONNET: {
+    provider: "anthropic",
+    providerModelId: "claude-sonnet-4-5",
+    aliases: ["sonnet"],
+    warm: true,
+    isDefault: true,
+    tokenizerOverride: "anthropic/claude-sonnet-4.5",
+  },
+  HAIKU: {
+    provider: "anthropic",
+    providerModelId: "claude-haiku-4-5",
+    aliases: ["haiku"],
+    tokenizerOverride: "anthropic/claude-3.5-haiku",
+  },
+  OPUS: {
+    provider: "anthropic",
+    providerModelId: "claude-opus-4-1",
+    aliases: ["opus"],
+  },
+  GPT: {
+    provider: "openai",
+    providerModelId: "gpt-5.1",
+    aliases: ["gpt-5.1"],
+    warm: true,
+  },
+  GPT_PRO: {
+    provider: "openai",
+    providerModelId: "gpt-5-pro",
+    aliases: ["gpt-5-pro"],
+  },
+  GPT_CODEX: {
+    provider: "openai",
+    providerModelId: "gpt-5.1-codex",
+    aliases: ["codex"],
+    warm: true,
+  },
+  GPT_MINI: {
+    provider: "openai",
+    providerModelId: "gpt-5.1-codex-mini",
+  },
+} as const satisfies Record<string, KnownModelDefinition>;
+
+export type KnownModelKey = keyof typeof MODEL_DEFINITIONS;
+
+export const KNOWN_MODELS = Object.fromEntries(
+  Object.entries(MODEL_DEFINITIONS).map(([key, definition]) => [
+    key,
+    {
+      ...definition,
+      id: `${definition.provider}:${definition.providerModelId}` as `${ModelProvider}:${string}`,
+    },
+  ])
+) as Record<KnownModelKey, KnownModel>;
+
+export function getKnownModel(key: KnownModelKey): KnownModel {
+  return KNOWN_MODELS[key];
+}
+
+// ------------------------------------------------------------------------------------
+// Derived collections
+// ------------------------------------------------------------------------------------
+
+const DEFAULT_MODEL_ENTRY =
+  Object.values(KNOWN_MODELS).find((model) => model.isDefault) ?? KNOWN_MODELS.SONNET;
+
+export const DEFAULT_MODEL = DEFAULT_MODEL_ENTRY.id;
+
+export const DEFAULT_WARM_MODELS = Object.values(KNOWN_MODELS)
+  .filter((model) => model.warm)
+  .map((model) => model.id);
+
+export const MODEL_ABBREVIATIONS: Record<string, string> = Object.fromEntries(
+  Object.values(KNOWN_MODELS)
+    .flatMap((model) => (model.aliases ?? []).map((alias) => [alias, model.id] as const))
+    .sort(([a], [b]) => a.localeCompare(b))
+);
+
+export const TOKENIZER_MODEL_OVERRIDES: Record<string, string> = Object.fromEntries(
+  Object.values(KNOWN_MODELS)
+    .filter((model) => Boolean(model.tokenizerOverride))
+    .map((model) => [model.id, model.tokenizerOverride!])
+);
+
+export const MODEL_NAMES: Record<ModelProvider, Record<string, string>> = Object.entries(
+  KNOWN_MODELS
+).reduce<Record<ModelProvider, Record<string, string>>>(
+  (acc, [key, model]) => {
+    if (!acc[model.provider]) {
+      const emptyRecord: Record<string, string> = {};
+      acc[model.provider] = emptyRecord;
+    }
+    acc[model.provider][key] = model.providerModelId;
+    return acc;
+  },
+  {} as Record<ModelProvider, Record<string, string>>
+);

src/hooks/useModelLRU.ts

@@ -7,7 +7,11 @@ const MAX_LRU_SIZE = 8;
 const LRU_KEY = "model-lru";
 
 // Default models from abbreviations (for initial LRU population)
-const DEFAULT_MODELS = Object.values(MODEL_ABBREVIATIONS);
+// Ensure defaultModel is first, then fill with other abbreviations
+const DEFAULT_MODELS = [
+  defaultModel,
+  ...Object.values(MODEL_ABBREVIATIONS).filter((m) => m !== defaultModel),
+].slice(0, MAX_LRU_SIZE);
 
 /**
  * Get the default model from LRU (non-hook version for use outside React)

src/services/historyService.ts

@@ -7,6 +7,7 @@ import type { Config } from "@/config";
 import { workspaceFileLocks } from "@/utils/concurrency/workspaceFileLocks";
 import { log } from "./log";
 import { getTokenizerForModel } from "@/utils/main/tokenizer";
+import { KNOWN_MODELS } from "@/constants/knownModels";
 
 /**
  * HistoryService - Manages chat history persistence and sequence numbering
@@ -340,7 +341,7 @@ export class HistoryService {
         }
 
         // Get tokenizer for counting (use a default model)
-        const tokenizer = await getTokenizerForModel("anthropic:claude-sonnet-4-5");
+        const tokenizer = await getTokenizerForModel(KNOWN_MODELS.SONNET.id);
 
         // Count tokens for each message
         // We stringify the entire message for simplicity - only relative weights matter

src/services/mock/mockScenarioPlayer.ts

@@ -18,8 +18,9 @@ import type { StreamStartEvent, StreamDeltaEvent, StreamEndEvent } from "@/types
 import type { ToolCallStartEvent, ToolCallEndEvent } from "@/types/stream";
 import type { ReasoningDeltaEvent } from "@/types/stream";
 import { getTokenizerForModel } from "@/utils/main/tokenizer";
+import { KNOWN_MODELS } from "@/constants/knownModels";
 
-const MOCK_TOKENIZER_MODEL = "openai:gpt-5";
+const MOCK_TOKENIZER_MODEL = KNOWN_MODELS.GPT.id;
 const TOKENIZE_TIMEOUT_MS = 150;
 let tokenizerFallbackLogged = false;
 

src/services/mock/scenarios/basicChat.ts

@@ -1,5 +1,6 @@
 import type { ScenarioTurn } from "../scenarioTypes";
 import { STREAM_BASE_DELAY } from "../scenarioTypes";
+import { KNOWN_MODELS } from "@/constants/knownModels";
 
 export const LIST_PROGRAMMING_LANGUAGES = "List 3 programming languages";
 
@@ -12,7 +13,7 @@ const listProgrammingLanguagesTurn: ScenarioTurn = {
   assistant: {
     messageId: "msg-basic-1",
     events: [
-      { kind: "stream-start", delay: 0, messageId: "msg-basic-1", model: "openai:gpt-5" },
+      { kind: "stream-start", delay: 0, messageId: "msg-basic-1", model: KNOWN_MODELS.GPT.id },
       {
         kind: "stream-delta",
         delay: STREAM_BASE_DELAY,
@@ -37,7 +38,7 @@ const listProgrammingLanguagesTurn: ScenarioTurn = {
         kind: "stream-end",
         delay: STREAM_BASE_DELAY * 5,
         metadata: {
-          model: "openai:gpt-5",
+          model: KNOWN_MODELS.GPT.id,
           inputTokens: 64,
           outputTokens: 48,
           systemMessageTokens: 12,

src/services/mock/scenarios/permissionModes.ts

@@ -1,4 +1,5 @@
 import type { ScenarioTurn } from "../scenarioTypes";
+import { KNOWN_MODELS } from "@/constants/knownModels";
 import { STREAM_BASE_DELAY } from "../scenarioTypes";
 
 export const PERMISSION_MODE_PROMPTS = {
@@ -19,7 +20,7 @@ const planRefactorTurn: ScenarioTurn = {
         kind: "stream-start",
         delay: 0,
         messageId: "msg-plan-refactor",
-        model: "openai:gpt-5",
+        model: KNOWN_MODELS.GPT.id,
       },
       {
         kind: "stream-delta",
@@ -45,7 +46,7 @@ const planRefactorTurn: ScenarioTurn = {
         kind: "stream-end",
         delay: STREAM_BASE_DELAY * 5,
         metadata: {
-          model: "openai:gpt-5",
+          model: KNOWN_MODELS.GPT.id,
           inputTokens: 180,
           outputTokens: 130,
           systemMessageTokens: 24,
@@ -74,7 +75,7 @@ const executePlanTurn: ScenarioTurn = {
         kind: "stream-start",
         delay: 0,
         messageId: "msg-exec-refactor",
-        model: "openai:gpt-5",
+        model: KNOWN_MODELS.GPT.id,
       },
       {
         kind: "tool-start",
@@ -118,7 +119,7 @@ const executePlanTurn: ScenarioTurn = {
         kind: "stream-end",
         delay: STREAM_BASE_DELAY * 3,
         metadata: {
-          model: "openai:gpt-5",
+          model: KNOWN_MODELS.GPT.id,
           inputTokens: 220,
           outputTokens: 110,
           systemMessageTokens: 18,

src/services/mock/scenarios/review.ts

@@ -1,4 +1,5 @@
 import type { ScenarioTurn } from "../scenarioTypes";
+import { KNOWN_MODELS } from "@/constants/knownModels";
 import { STREAM_BASE_DELAY } from "../scenarioTypes";
 
 export const REVIEW_PROMPTS = {
@@ -16,7 +17,7 @@ const summarizeBranchesTurn: ScenarioTurn = {
   assistant: {
     messageId: "msg-plan-1",
     events: [
-      { kind: "stream-start", delay: 0, messageId: "msg-plan-1", model: "openai:gpt-5" },
+      { kind: "stream-start", delay: 0, messageId: "msg-plan-1", model: KNOWN_MODELS.GPT.id },
       {
         kind: "reasoning-delta",
         delay: STREAM_BASE_DELAY,
@@ -61,7 +62,7 @@ const summarizeBranchesTurn: ScenarioTurn = {
         kind: "stream-end",
         delay: STREAM_BASE_DELAY * 6,
         metadata: {
-          model: "openai:gpt-5",
+          model: KNOWN_MODELS.GPT.id,
           inputTokens: 128,
           outputTokens: 85,
           systemMessageTokens: 32,
@@ -86,7 +87,7 @@ const openOnboardingDocTurn: ScenarioTurn = {
   assistant: {
     messageId: "msg-exec-1",
     events: [
-      { kind: "stream-start", delay: 0, messageId: "msg-exec-1", model: "openai:gpt-5" },
+      { kind: "stream-start", delay: 0, messageId: "msg-exec-1", model: KNOWN_MODELS.GPT.id },
       {
         kind: "tool-start",
         delay: STREAM_BASE_DELAY,
@@ -114,7 +115,7 @@ const showOnboardingDocTurn: ScenarioTurn = {
   assistant: {
     messageId: "msg-exec-2",
     events: [
-      { kind: "stream-start", delay: 0, messageId: "msg-exec-2", model: "openai:gpt-5" },
+      { kind: "stream-start", delay: 0, messageId: "msg-exec-2", model: KNOWN_MODELS.GPT.id },
       {
         kind: "tool-start",
         delay: STREAM_BASE_DELAY,
@@ -153,7 +154,7 @@ const showOnboardingDocTurn: ScenarioTurn = {
         kind: "stream-end",
         delay: STREAM_BASE_DELAY * 3,
         metadata: {
-          model: "openai:gpt-5",
+          model: KNOWN_MODELS.GPT.id,
           inputTokens: 96,
           outputTokens: 142,
           systemMessageTokens: 32,

src/services/mock/scenarios/slashCommands.ts

@@ -1,4 +1,5 @@
 import type { ScenarioTurn } from "../scenarioTypes";
+import { KNOWN_MODELS } from "@/constants/knownModels";
 import { STREAM_BASE_DELAY } from "../scenarioTypes";
 
 export const SLASH_COMMAND_PROMPTS = {
@@ -24,7 +25,7 @@ const compactConversationTurn: ScenarioTurn = {
         kind: "stream-start",
         delay: 0,
         messageId: "msg-slash-compact-1",
-        model: "openai:gpt-5",
+        model: KNOWN_MODELS.GPT.id,
       },
       {
         kind: "stream-delta",
@@ -35,7 +36,7 @@ const compactConversationTurn: ScenarioTurn = {
         kind: "stream-end",
         delay: STREAM_BASE_DELAY * 2,
         metadata: {
-          model: "openai:gpt-5",
+          model: KNOWN_MODELS.GPT.id,
           inputTokens: 220,
           outputTokens: 96,
           systemMessageTokens: 18,