feat(providers): adopt gpt-5.6 model family and new codex effort tiers

docs/features/command-palette.md

@@ -88,8 +88,8 @@ This is usually faster than opening Settings and browsing manually.
 Default bindings:
 
 - `Alt+1` → `Claude Opus 4.8`
-- `Alt+2` → `GPT-5.5`
-- `Alt+3` → `GPT-5.4`
+- `Alt+2` → `GPT-5.6 Terra`
+- `Alt+3` → `GPT-5.6 Sol`
 
 ### Work Mostly From The Keyboard
 

docs/features/workspace-latest-turn-summary.md

@@ -64,7 +64,7 @@ This rendered example shows the Information panel with the latest-turn summary p
 
 ```json
 {
-  "workspaceTurnSummaryPrimaryModel": "gpt-5.4-mini",
+  "workspaceTurnSummaryPrimaryModel": "gpt-5.6-luna",
   "workspaceTurnSummaryFallbackModel": "claude-haiku-4-5"
 }
 ```

docs/providers/provider-runtimes.md

@@ -251,12 +251,15 @@ If you want the user-facing setup workflow instead of the runtime internals, use
   - `untrusted`: App Server-aligned low-friction default; pause only for actions treated as untrusted.
   - `on-request`: ask when approval is needed.
 - `reasoning effort`
-  - `minimal` / `low`: fastest.
+  - `low`: fastest.
   - `medium`: balanced default.
   - `high` / `xhigh`: slower, more deliberate.
-  - Codex App Server turns normalize `minimal` to `low` because the current
-    upstream model API rejects built-in tools such as `image_gen` and
-    `web_search` with `reasoning.effort = minimal`.
+  - `max` / `ultra`: deepest reasoning tiers introduced with the GPT-5.6
+    Codex CLI scale.
+  - `minimal` is a legacy persisted value that is no longer selectable in the
+    UI; Codex App Server turns normalize it to `low` because the upstream
+    model API rejects built-in tools such as `image_gen` and `web_search`
+    with `reasoning.effort = minimal`.
 - `reasoning summary`
   - `auto`: let Codex decide.
   - `concise`: short summary.
@@ -283,7 +286,7 @@ When a task switches from one Codex model to another, Stave does not attempt to
 
 - Codex App Server transport: local `codex app-server` from Codex CLI `0.142.0`
 - Codex CLI baseline: `0.142.0`
-- Current Stave-supported Codex model IDs: `gpt-5.5`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.3-codex-spark`
+- Current Stave-supported Codex model IDs: `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5` (default: `gpt-5.6-terra`)
 
 Stave requires a user-installed Codex CLI (`codex` ≥ 0.142.0). Users must have Codex CLI available in their PATH or configured via `runtimeOptions.codexBinaryPath` / `STAVE_CODEX_CLI_PATH`. A user-configured binary path still takes precedence over auto-discovery.
 

electron/main/ipc/schemas.ts

@@ -392,6 +392,8 @@ export const RuntimeOptionsObjectSchema = z
         z.literal("medium"),
         z.literal("high"),
         z.literal("xhigh"),
+        z.literal("max"),
+        z.literal("ultra"),
       ])
       .optional(),
     codexWebSearch: z

src/components/ai-elements/prompt-input.tsx

@@ -300,7 +300,7 @@ function getPromptToolbarAccentClass(
 }
 
 function isHighestEffortValue(value?: string) {
-  return value === "max" || value === "xhigh";
+  return value === "ultra" || value === "max" || value === "xhigh";
 }
 
 function getEffortIconToneClass(value?: string) {

src/components/session/chat-input.runtime.ts

@@ -70,12 +70,15 @@ type CommandCatalogRuntimeArgs = Pick<
 const CLAUDE_EFFORT_CYCLE_ORDER = CLAUDE_EFFORT_OPTIONS.map(
   (option) => option.value,
 );
+// Legacy "minimal" is intentionally absent — cycling from a persisted
+// "minimal" value falls back to the first entry ("low").
 const CODEX_EFFORT_CYCLE_ORDER = [
   "low",
   "medium",
   "high",
   "xhigh",
-  "minimal",
+  "max",
+  "ultra",
 ] as const satisfies readonly AppSettings["codexReasoningEffort"][];
 
 function cycleOptionValue<T extends string>(args: {

src/lib/providers/codex-runtime-options.ts

@@ -55,7 +55,9 @@ export function resolveCodexAppServerReasoningEffort(args: {
     args.reasoningEffort !== "low" &&
     args.reasoningEffort !== "medium" &&
     args.reasoningEffort !== "high" &&
-    args.reasoningEffort !== "xhigh"
+    args.reasoningEffort !== "xhigh" &&
+    args.reasoningEffort !== "max" &&
+    args.reasoningEffort !== "ultra"
   ) {
     return undefined;
   }

src/lib/providers/model-catalog.ts

@@ -42,12 +42,15 @@ export const CLAUDE_SDK_MODEL_OPTIONS = [
 
 // Source:
 // - local `codex app-server` / CLI baseline support
-// - https://developers.openai.com/codex/models
+// - https://developers.openai.com/codex/models (GPT-5.6 family, 2026-07-09)
+// GPT-5.6 ships as Sol (flagship), Terra (balanced), and Luna (fast/cheap).
+// gpt-5.5 is kept as the previous-generation fallback; gpt-5.4 variants and
+// codex-spark moved to legacy display names only.
 export const CODEX_MODEL_OPTIONS = [
+  "gpt-5.6-sol",
+  "gpt-5.6-terra",
+  "gpt-5.6-luna",
   "gpt-5.5",
-  "gpt-5.4",
-  "gpt-5.4-mini",
-  "gpt-5.3-codex-spark",
 ] as const;
 
 export interface ProviderDescriptor {
@@ -87,7 +90,7 @@ export const PROVIDER_DESCRIPTORS = [
     iconUrl: CODEX_COLOR_ICON_URL,
     fallbackLabel: "O",
     models: CODEX_MODEL_OPTIONS,
-    defaultModel: "gpt-5.5",
+    defaultModel: "gpt-5.6-terra",
     sessionLabel: "Codex thread ID",
     capabilities: {
       nativeCommandCatalog: true,
@@ -227,14 +230,6 @@ export const MODEL_TIER_ORDER = [
   "frontier",
 ] as const satisfies readonly ModelTier[];
 
-const GENERAL_CODING_TASK_TYPES = [
-  "quick_edit",
-  "implementation",
-  "debug",
-  "review",
-  "general",
-] as const satisfies readonly TaskType[];
-
 export const MODEL_CAPABILITIES: Record<string, ModelCapability> = {
   [CLAUDE_FABLE_MODEL]: {
     providerId: "claude-code",
@@ -285,32 +280,32 @@ export const MODEL_CAPABILITIES: Record<string, ModelCapability> = {
     taskTypes: ["quick_edit", "general"],
     defaultClaudeEffort: "medium",
   },
-  "gpt-5.5": {
+  "gpt-5.6-sol": {
     providerId: "codex",
-    model: "gpt-5.5",
+    model: "gpt-5.6-sol",
     tier: "frontier",
     taskTypes: ["plan", "implementation", "debug", "review", "safety"],
     defaultCodexReasoningEffort: "high",
   },
-  "gpt-5.4": {
+  "gpt-5.6-terra": {
     providerId: "codex",
-    model: "gpt-5.4",
-    tier: "standard",
-    taskTypes: GENERAL_CODING_TASK_TYPES,
+    model: "gpt-5.6-terra",
+    tier: "heavy",
+    taskTypes: ["plan", "implementation", "debug", "review", "safety"],
     defaultCodexReasoningEffort: "medium",
   },
-  "gpt-5.4-mini": {
+  "gpt-5.6-luna": {
     providerId: "codex",
-    model: "gpt-5.4-mini",
+    model: "gpt-5.6-luna",
     tier: "light",
     taskTypes: ["quick_edit", "general"],
     defaultCodexReasoningEffort: "low",
   },
-  "gpt-5.3-codex-spark": {
+  "gpt-5.5": {
     providerId: "codex",
-    model: "gpt-5.3-codex-spark",
-    tier: "heavy",
-    taskTypes: ["implementation", "debug", "review"],
+    model: "gpt-5.5",
+    tier: "frontier",
+    taskTypes: ["plan", "implementation", "debug", "review", "safety"],
     defaultCodexReasoningEffort: "high",
   },
 };
@@ -419,10 +414,14 @@ export function resolveDefaultCodexEffortForModel(args: {
   }
 
   const normalizedModel = args.model.trim().toLowerCase();
-  if (normalizedModel.includes("mini")) {
+  if (normalizedModel.includes("mini") || normalizedModel.includes("luna")) {
     return "low";
   }
-  if (normalizedModel.includes("spark") || normalizedModel.includes("5.5")) {
+  if (
+    normalizedModel.includes("spark") ||
+    normalizedModel.includes("sol") ||
+    normalizedModel.includes("5.5")
+  ) {
     return "high";
   }
   return "medium";
@@ -492,7 +491,12 @@ export function toHumanModelName(args: { model: string }) {
     "claude-sonnet-4-6": "Claude Sonnet 4.6",
     "claude-sonnet-4-6[1m]": "Claude Sonnet 4.6 (1M)",
     "claude-haiku-4-5": "Claude Haiku 4.5",
+    "gpt-5.6-sol": "GPT-5.6 Sol",
+    "gpt-5.6-terra": "GPT-5.6 Terra",
+    "gpt-5.6-luna": "GPT-5.6 Luna",
     "gpt-5.5": "GPT-5.5",
+    // Legacy Codex labels kept so historical chat/turn records still render
+    // recognizable names after the picker lineup moved to GPT-5.6.
     "gpt-5.4": "GPT-5.4",
     "gpt-5.4-mini": "GPT-5.4 Mini",
     "gpt-5-codex": "GPT-5-Codex",

src/lib/providers/model-shortcuts.ts

@@ -20,8 +20,8 @@ export const MODEL_SHORTCUT_SLOT_LABELS = [
 
 export const DEFAULT_MODEL_SHORTCUT_KEYS = [
   "claude-code:claude-opus-4-8",
-  "codex:gpt-5.5",
-  "codex:gpt-5.4",
+  "codex:gpt-5.6-terra",
+  "codex:gpt-5.6-sol",
   "",
   "",
   "",

src/lib/providers/provider.types.ts

@@ -582,7 +582,16 @@ export interface ProviderRuntimeOptions {
   codexNetworkAccess?: boolean;
   codexApprovalPolicy?: "never" | "on-request" | "on-failure" | "untrusted";
   codexBinaryPath?: string;
-  codexReasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh";
+  // "minimal" is a legacy value kept for persisted settings; the runtime maps
+  // it to "low". "max" and "ultra" arrived with the GPT-5.6 Codex CLI scale.
+  codexReasoningEffort?:
+    | "minimal"
+    | "low"
+    | "medium"
+    | "high"
+    | "xhigh"
+    | "max"
+    | "ultra";
   codexWebSearch?: "disabled" | "cached" | "live";
   codexShowRawReasoning?: boolean;
   codexReasoningSummary?: "auto" | "concise" | "detailed" | "none";

src/lib/providers/runtime-option-contract.ts

@@ -65,12 +65,16 @@ export const CODEX_SANDBOX_MODE_OPTIONS = [
   { value: "danger-full-access", label: "danger-full-access" },
 ] as const satisfies readonly SelectOption<NonNullable<ProviderRuntimeOptions["codexFileAccess"]>>[];
 
+// "minimal" was dropped from the Codex CLI effort scale with GPT-5.6 and is
+// no longer selectable; legacy persisted values still validate and map to
+// "low" at runtime (see resolveCodexAppServerReasoningEffort).
 export const CODEX_EFFORT_OPTIONS = [
-  { value: "minimal", label: "Minimal" },
   { value: "low", label: "Low" },
   { value: "medium", label: "Medium" },
   { value: "high", label: "High" },
   { value: "xhigh", label: "X-High" },
+  { value: "max", label: "Max" },
+  { value: "ultra", label: "Ultra" },
 ] as const satisfies readonly SelectOption<NonNullable<ProviderRuntimeOptions["codexReasoningEffort"]>>[];
 
 export const CODEX_WEB_SEARCH_OPTIONS = [

src/lib/task-presets.ts

@@ -30,7 +30,8 @@ export interface TaskPreset {
   /**
    * Reasoning effort applied to `task` presets. Claude presets accept
    * `low | medium | high | xhigh | max`; Codex presets accept
-   * `minimal | low | medium | high | xhigh`. When omitted, the model's
+   * `low | medium | high | xhigh | max | ultra` (legacy persisted `minimal`
+   * still validates and maps to `low` at runtime). When omitted, the model's
    * default effort is used at launch. Ignored for CLI sessions.
    */
   effort?: TaskPresetEffort;
@@ -46,7 +47,8 @@ export type TaskPresetEffort =
   | "medium"
   | "high"
   | "xhigh"
-  | "max";
+  | "max"
+  | "ultra";
 
 /**
  * Effort options selectable for a `task` preset, scoped to the provider's
@@ -91,11 +93,11 @@ export const DEFAULT_TASK_PRESETS: readonly TaskPreset[] = [
     model: DEFAULT_CLAUDE_OPUS_MODEL,
   },
   {
-    id: "default-gpt-5-5-task",
-    label: "GPT-5.5",
+    id: "default-gpt-5-6-task",
+    label: "GPT-5.6",
     kind: "task",
     provider: "codex",
-    model: "gpt-5.5",
+    model: "gpt-5.6-terra",
   },
   {
     id: "default-claude-cli-session",

src/store/app.store.ts

@@ -1470,7 +1470,14 @@ export interface AppSettings {
   codexNetworkAccess: boolean;
   codexApprovalPolicy: "never" | "on-request" | "on-failure" | "untrusted";
   codexBinaryPath: string;
-  codexReasoningEffort: "minimal" | "low" | "medium" | "high" | "xhigh";
+  codexReasoningEffort:
+    | "minimal"
+    | "low"
+    | "medium"
+    | "high"
+    | "xhigh"
+    | "max"
+    | "ultra";
   codexWebSearch: "disabled" | "cached" | "live";
   codexShowRawReasoning: boolean;
   codexReasoningSummary: "auto" | "concise" | "detailed" | "none";
@@ -2675,7 +2682,7 @@ const defaultSettings: AppSettings = {
   promptResponseStyle: DEFAULT_PROMPT_RESPONSE_STYLE,
   promptPrDescription: DEFAULT_PROMPT_PR_DESCRIPTION,
   promptInlineCompletion: DEFAULT_PROMPT_INLINE_COMPLETION,
-  workspaceTurnSummaryPrimaryModel: "gpt-5.4-mini",
+  workspaceTurnSummaryPrimaryModel: "gpt-5.6-luna",
   workspaceTurnSummaryFallbackModel: "claude-haiku-4-5",
   workspaceTurnSummaryPrompt: DEFAULT_PROMPT_WORKSPACE_TURN_SUMMARY,
 

tests/chat-input.runtime.test.ts

@@ -45,11 +45,15 @@ describe("chat-input runtime helpers", () => {
     expect(cycleClaudeEffortValue("max")).toBe("low");
   });
 
-  test("cycles Codex effort with minimal at the end of the loop", () => {
+  test("cycles Codex effort through max and ultra, mapping legacy minimal to low", () => {
     expect(cycleCodexEffortValue("low")).toBe("medium");
     expect(cycleCodexEffortValue("medium")).toBe("high");
     expect(cycleCodexEffortValue("high")).toBe("xhigh");
-    expect(cycleCodexEffortValue("xhigh")).toBe("minimal");
+    expect(cycleCodexEffortValue("xhigh")).toBe("max");
+    expect(cycleCodexEffortValue("max")).toBe("ultra");
+    expect(cycleCodexEffortValue("ultra")).toBe("low");
+    // Legacy persisted "minimal" is no longer in the cycle; it falls back to
+    // the first entry.
     expect(cycleCodexEffortValue("minimal")).toBe("low");
   });
 

tests/codex-runtime-options.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect, test } from "bun:test";
 import {
+  resolveCodexAppServerReasoningEffort,
   resolveEffectiveCodexApprovalPolicy,
   resolveEffectiveCodexFileAccessMode,
 } from "@/lib/providers/codex-runtime-options";
@@ -42,3 +43,26 @@ describe("resolveEffectiveCodexApprovalPolicy", () => {
     })).toBe("untrusted");
   });
 });
+
+describe("resolveCodexAppServerReasoningEffort", () => {
+  test("passes through the GPT-5.6 max and ultra effort tiers", () => {
+    expect(
+      resolveCodexAppServerReasoningEffort({ reasoningEffort: "max" }),
+    ).toBe("max");
+    expect(
+      resolveCodexAppServerReasoningEffort({ reasoningEffort: "ultra" }),
+    ).toBe("ultra");
+  });
+
+  test("normalizes legacy minimal to low", () => {
+    expect(
+      resolveCodexAppServerReasoningEffort({ reasoningEffort: "minimal" }),
+    ).toBe("low");
+  });
+
+  test("returns undefined for unknown or missing efforts", () => {
+    expect(
+      resolveCodexAppServerReasoningEffort({ reasoningEffort: undefined }),
+    ).toBeUndefined();
+  });
+});

tests/ipc-schemas.test.ts

@@ -45,6 +45,20 @@ describe("provider IPC schemas", () => {
     expect(parsed.success).toBe(true);
   });
 
+  test("accepts Codex max/ultra efforts and legacy minimal in runtime options", () => {
+    for (const effort of ["max", "ultra", "minimal"] as const) {
+      const parsed = StreamTurnArgsSchema.safeParse({
+        providerId: "codex",
+        prompt: "continue",
+        runtimeOptions: {
+          codexReasoningEffort: effort,
+        },
+      });
+
+      expect(parsed.success).toBe(true);
+    }
+  });
+
   test("accepts provider timeout windows up to 24 hours", () => {
     expect(StreamTurnArgsSchema.safeParse({
       providerId: "codex",