fix(ai): recognise the gpt-6 model family in every model gate

The ChatGPT Codex backend and the OpenAI API serve gpt-6-sol, gpt-6-luna,
gpt-6-astra and gpt-6.1-sol, but the reasoning, Chat Completions and vision
gates only knew the gpt-5 line. A gpt-6 model got no reasoning control and
no real "off", went out on the legacy wire with max_tokens and a
temperature that reasoning models refuse, and was denied image input.

The three gates now read one classifier, openai-model-family.ts, so a new
generation is taught in one place. Every GPT from 5.1 on, the gpt-6 line
included, is the family that takes effort "none"; every GPT from 5 on reads
images. The upstream catalog (codex-rs/models-manager/models.json) lists
image input for every gpt-6 slug and the same reasoning presets as the
gpt-5.6 line; the presets omit "none" for gpt-5.5 too, which accepts it.

The Codex slug ladder moves into codex-slug-chain.ts and gains gpt-6-sol,
gpt-6.1-sol and gpt-6-luna. gpt-5.6-terra stays first as the slug verified
against the backend; the catalog names gpt-6-sol as its upgrade, and a
rotation retires a whole line at once, so the gpt-6 rungs follow before
terra's own siblings. The ladder test now holds every rung to a recognised
reasoning family with a real off and image input. The OpenAI presets offer
gpt-6.1-sol and gpt-6-luna.

docs/codex-protocol-spec.md

@@ -630,7 +630,8 @@ Confirmed-rejected slugs (despite the source-tree references in the table above)
 
 Confirmed-accepted slugs on a ChatGPT Plus subscription (per `available_in_plans` in `models.json`):
 
-- `gpt-5.6-terra` — HealthLog's default (first chain entry) as of 2026-10-24, with `gpt-5.6-luna` and `gpt-5.6-sol` behind it; the backend rejected `gpt-5.5` and everything below it that day
+- `gpt-5.6-terra` — HealthLog's default (first chain entry) as of 2026-10-24; the backend rejected `gpt-5.5` and everything below it that day. `gpt-5.6-luna` and `gpt-5.6-sol` are its siblings on the same line
+- `gpt-6-sol`, `gpt-6.1-sol`, `gpt-6-luna`, `gpt-6-astra` — the gpt-6 line, listed in `models.json` since 2026-09-22 (`gpt-6.1-sol` the catalog default since 2026-09-29). The catalog names `gpt-6-sol` as the upgrade of `gpt-5.6-terra` and `gpt-5.6-sol`, and `gpt-6-luna` as that of `gpt-5.6-luna`. Every gpt-6 slug lists `input_modalities: ["text", "image"]` and the same reasoning presets as the gpt-5.6 line (`low` to `max`, sol and astra also `ultra`). The presets omit `none`, as they do for `gpt-5.5`, which a live probe on 2026-10-06 showed accepts it; HealthLog sends `none` for "off" on the gpt-6 line too and learns a refusal per slug. The gpt-6 slugs were not probed against a live backend when they joined the chain
 - `gpt-5.5` — the default from 2026-06 until 2026-10-24
 - `gpt-5.4` / `gpt-5.4-mini` / `gpt-5.2` — accepted per plan
 - `gpt-5.3-codex` — was the default (verified 2026-05-09); the backend rotated off the `-codex` line on 2026-06-02 and this slug is now only accepted intermittently, so it sits on a late chain rung
@@ -643,18 +644,23 @@ The ChatGPT-Codex allow-list is rotated by OpenAI without prior notice (`gpt-5-c
 
 To eliminate that risk class, HealthLog's `CodexClient` walks an ordered chain on every fresh request series.
 
-**Chain** (top = first try; default order is the empirically-safest most-current-first ordering as of 2026-06, after the backend rotated off the `-codex` line on 2026-06-02):
+**Chain** (top = first try). `gpt-5.6-terra` leads because it is the slug verified against the backend. The gpt-6 rungs follow before terra's own siblings: the catalog names `gpt-6-sol` as terra's upgrade, and the 2026-10-24 rotation retired a whole line at once, so a rotation that takes terra most likely takes `gpt-5.6-luna` and `gpt-5.6-sol` with it. `gpt-6.1-sol` (the catalog default, also on plans `gpt-6-sol` is not offered on) and `gpt-6-luna` (the lighter one) come next. `gpt-6-astra`, the frontier tier, stays out of the default:
 
 ```ts
 const DEFAULT_SLUG_FALLBACK_CHAIN = [
-  "gpt-5.6-terra", // current default, verified against the backend 2026-10-24
-  "gpt-5.6-luna", // lighter sibling on the same line
-  "gpt-5.6-sol", // heavier sibling on the same line
+  "gpt-5.6-terra", // verified against the backend 2026-10-24
+  "gpt-6-sol", // terra's upgrade in the upstream catalog
+  "gpt-6.1-sol", // the catalog default, on every plan
+  "gpt-6-luna", // lighter gpt-6 sibling
+  "gpt-5.6-luna", // lighter sibling on terra's line
+  "gpt-5.6-sol", // heavier sibling on terra's line
   "gpt-5.5", // previous default, kept as the last rung while plans differ
 ];
 ```
 
-The shipped constant in `src/lib/ai/codex-client.ts` is authoritative; when the upstream rotates again, that file moves first and this listing follows.
+Every default rung must be a model family the reasoning, sampling and vision gates recognise (`src/lib/ai/openai-model-family.ts`); `codex-slug-fallback.test.ts` holds the chain to that.
+
+The shipped constant in `src/lib/ai/codex-slug-chain.ts` is authoritative; when the upstream rotates again, that file moves first and this listing follows.
 
 The chain is overridable via the env var `CODEX_MODEL_FALLBACK_CHAIN` (comma-separated). The first env-var-named slug supersedes `CODEX_MODEL`; if both are set, `CODEX_MODEL` is folded into position 0 and de-duplicated.
 

src/__tests__/baselines/source-file-lines.json

@@ -1,7 +1,7 @@
 {
   "$comment": "Production files over 800 lines and their length. Maintained by src/__tests__/source-size-ratchet.test.ts.",
   "count": 89,
-  "total": 99565,
+  "total": 99521,
   "entries": {
     "app/medications/page-client.tsx": 839,
     "components/charts/mood-chart.tsx": 1166,
@@ -29,7 +29,7 @@
     "lib/ai/coach/stream-events.ts": 817,
     "lib/ai/coach/turn/model.ts": 820,
     "lib/ai/coach/types.ts": 1011,
-    "lib/ai/codex-client.ts": 1089,
+    "lib/ai/codex-client.ts": 1045,
     "lib/ai/local-client.ts": 880,
     "lib/ai/prompts/insight-generator.ts": 1110,
     "lib/ai/provider.ts": 1632,

src/components/settings/ai/shared.ts

@@ -99,10 +99,13 @@ export const DEFAULT_CHAIN: readonly ChainEntry[] = [
  * OpenAI model presets — the wire-up keeps users out of the freetext
  * box for the common case. The `__custom__` sentinel surfaces a
  * freetext input so power users can target preview models or
- * fine-tunes (e.g. `gpt-5`, which is admitted via the env override on
- * the Codex backend but not yet a documented OpenAI model id).
+ * fine-tunes. The gpt-6 entries are the current workhorse and its
+ * lighter sibling; every preset is a family `openai-model-family.ts`
+ * recognises, so the reasoning, token-field and vision gates fit it.
  */
 export const OPENAI_MODEL_PRESETS = [
+  "gpt-6.1-sol",
+  "gpt-6-luna",
   "gpt-4o",
   "gpt-4o-mini",
   "gpt-4-turbo",

src/lib/ai/__tests__/codex-slug-fallback.test.ts

@@ -1,7 +1,12 @@
 import { describe, it, expect, vi, beforeEach, afterEach } from "vitest";
 import { CodexClient, __test } from "../codex-client";
 import { singleUserTurn } from "../types";
-import { openAIReasoningFamily } from "../reasoning/support";
+import {
+  openAIModelFamily,
+  type OpenAIReasoningFamily,
+} from "../openai-model-family";
+import { openAIOffEffort } from "../reasoning/support";
+import { supportsVisionForConfig } from "../vision-capability";
 import {
   clearCodexSlugCache,
   getCachedCodexSlug,
@@ -343,22 +348,39 @@ describe("Codex slug fallback chain", () => {
 
   it("DEFAULT_SLUG_FALLBACK_CHAIN tracks the current ChatGPT-auth ladder", () => {
     // The backend rejected every slug from gpt-5.5 down on 2026-10-24 and
-    // served the gpt-5.6 line; the in-code default must match the accepted
-    // set, in order.
+    // served the gpt-5.6 and gpt-6 lines. Terra, verified live, leads; its
+    // upstream successor and the rest of the gpt-6 line follow before terra's
+    // own siblings, which a line rotation would retire together with it.
     expect([...__test.DEFAULT_SLUG_FALLBACK_CHAIN]).toEqual([
       "gpt-5.6-terra",
+      "gpt-6-sol",
+      "gpt-6.1-sol",
+      "gpt-6-luna",
       "gpt-5.6-luna",
       "gpt-5.6-sol",
       "gpt-5.5",
     ]);
   });
 
-  it("DEFAULT_SLUG_FALLBACK_CHAIN holds only slugs the reasoning gate recognises", () => {
-    // A slug whose family the reasoning gate does not know goes out with the
-    // wrong parameters. A new line (gpt-6) joins the default only together
-    // with its family support.
+  it("DEFAULT_SLUG_FALLBACK_CHAIN holds only slugs every family gate recognises", () => {
+    // A slug whose family the gates do not know goes out with the wrong
+    // reasoning parameters and is denied image input. Every rung, whatever
+    // its generation, must be a recognised reasoning family with a real
+    // "off" (the Codex client sends `effort: "none"` for it) that reads
+    // images. Nothing is filtered out first: a new line joins the default
+    // only together with its family support.
+    expect(__test.DEFAULT_SLUG_FALLBACK_CHAIN.length).toBeGreaterThan(0);
     for (const slug of __test.DEFAULT_SLUG_FALLBACK_CHAIN) {
-      expect(openAIReasoningFamily(slug), slug).toBe("gpt-5.1+");
+      const family = openAIModelFamily(slug);
+      expect(family, slug).not.toBeNull();
+      expect(family?.reasoning, slug).not.toBeNull();
+      expect(
+        openAIOffEffort(
+          family!.reasoning as NonNullable<OpenAIReasoningFamily>,
+        ),
+        slug,
+      ).toBe("none");
+      expect(supportsVisionForConfig("codex", slug), slug).toBe(true);
     }
   });
 
@@ -370,6 +392,9 @@ describe("Codex slug fallback chain", () => {
     try {
       expect(__test.loadFallbackChain()).toEqual([
         "gpt-5.6-terra",
+        "gpt-6-sol",
+        "gpt-6.1-sol",
+        "gpt-6-luna",
         "gpt-5.6-luna",
         "gpt-5.6-sol",
         "gpt-5.5",

src/lib/ai/__tests__/openai-client.test.ts

@@ -157,6 +157,9 @@ describe("OpenAIClient", () => {
     ["o3-mini", 321, "https://api.openai.com/v1"],
     ["o4-mini-2025-04-16", 321, "https://api.openai.com/v1"],
     ["gpt-5.4-mini", 321, "https://API.OPENAI.COM:443/v1/"],
+    ["gpt-6-sol", 321, "https://api.openai.com/v1"],
+    ["gpt-6-luna", 1000, "https://api.openai.com/v1"],
+    ["gpt-6.1-sol", 321, "https://api.openai.com/v1"],
   ])(
     "uses the canonical modern token wire for %s",
     async (model, expectedBudget, baseUrl) => {

src/lib/ai/__tests__/openai-model-family.test.ts

@@ -0,0 +1,65 @@
+import { describe, expect, it } from "vitest";
+import { OPENAI_MODEL_PRESETS } from "@/components/settings/ai/shared";
+import { openAIModelFamily } from "../openai-model-family";
+
+describe("openAIModelFamily", () => {
+  it.each([
+    // The gpt-6 line as the upstream Codex catalog lists it.
+    ["gpt-6-sol", "gpt-5.1+", true],
+    ["gpt-6-luna", "gpt-5.1+", true],
+    ["gpt-6-astra", "gpt-5.1+", true],
+    ["gpt-6.1-sol", "gpt-5.1+", true],
+    ["gpt-6", "gpt-5.1+", true],
+    ["GPT-6.1-Sol", "gpt-5.1+", true],
+    ["openai/gpt-6-luna", "gpt-5.1+", true],
+    // A later generation reads like the newest one this file knows.
+    ["gpt-7", "gpt-5.1+", true],
+    // gpt-5.x
+    ["gpt-5.6-terra", "gpt-5.1+", true],
+    ["gpt-5.5", "gpt-5.1+", true],
+    ["gpt-5.3-codex", "gpt-5.1+", true],
+    ["gpt-5.1-codex-mini", "gpt-5.1+", true],
+    ["gpt-5", "gpt-5", true],
+    ["gpt-5-mini", "gpt-5", true],
+    ["gpt-5-codex", "gpt-5", true],
+    ["gpt-5-2025-08-07", "gpt-5", true],
+    // o-series: full models read images, -mini / -preview do not.
+    ["o1", "o", true],
+    ["o3", "o", true],
+    ["o4", "o", true],
+    ["o1-mini", "o", false],
+    ["o1-preview", "o", false],
+    ["o3-mini", "o", false],
+    ["o4-mini-2025-04-16", "o", false],
+    // gpt-4 class: no reasoning; vision on 4o, 4.1 and turbo only.
+    ["gpt-4o", null, true],
+    ["gpt-4o-mini", null, true],
+    ["gpt-4.1", null, true],
+    ["gpt-4.1-mini", null, true],
+    ["gpt-4-turbo", null, true],
+    ["gpt-4", null, false],
+    ["gpt-3.5-turbo", null, false],
+  ] as const)("%s", (model, reasoning, vision) => {
+    expect(openAIModelFamily(model)).toEqual({ reasoning, vision });
+  });
+
+  it("does not recognise names outside the OpenAI families", () => {
+    for (const model of [
+      "llama3.1:8b",
+      "claude-opus-5",
+      "gpt-daybreak-blue-latest",
+      "gpt-60x",
+      "o2",
+      "o3x",
+      "",
+    ]) {
+      expect(openAIModelFamily(model), model).toBeNull();
+    }
+  });
+
+  it("recognises every OpenAI model the settings offer as a preset", () => {
+    for (const model of OPENAI_MODEL_PRESETS) {
+      expect(openAIModelFamily(model), model).not.toBeNull();
+    }
+  });
+});

src/lib/ai/__tests__/openai-reasoning.test.ts

@@ -137,6 +137,24 @@ describe("OpenAI reasoning", () => {
     expect(body(mock, 2).reasoning_effort).toBe("low");
     expect(body(mock, 3)).not.toHaveProperty("reasoning_effort");
   });
+
+  it("controls reasoning on the gpt-6 line like the gpt-5.1+ line", async () => {
+    const mock = vi.fn().mockImplementation(plain("ok"));
+    vi.stubGlobal("fetch", mock);
+    await openai("gpt-6-sol").generateCompletion(
+      params({ reasoning: { effort: "off", summaries: false } }),
+    );
+    await openai("gpt-6.1-sol").generateCompletion(
+      params({
+        temperature: 0.3,
+        reasoning: { effort: "medium", summaries: false },
+      }),
+    );
+    expect(body(mock, 0).reasoning_effort).toBe("none");
+    expect(body(mock, 1).reasoning_effort).toBe("medium");
+    expect(body(mock, 1).max_completion_tokens).toBe(600 + 4096);
+    expect(body(mock, 1)).not.toHaveProperty("temperature");
+  });
 });
 
 describe("Gateway reasoning", () => {

src/lib/ai/__tests__/vision-capability.test.ts

@@ -5,6 +5,14 @@ import {
   supportsPdfForProvider,
 } from "../vision-capability";
 
+const GPT6_SLUGS = [
+  "gpt-6-sol",
+  "gpt-6-luna",
+  "gpt-6-astra",
+  "gpt-6.1-sol",
+  "gpt-6",
+] as const;
+
 describe("supportsVisionForConfig", () => {
   it("accepts the Claude 3 / 3.5 / 3.7 / 4 families for anthropic", () => {
     for (const model of [
@@ -60,6 +68,20 @@ describe("supportsVisionForConfig", () => {
     }
   });
 
+  it("accepts the gpt-6 line for openai + admin variants", () => {
+    // Every gpt-6 slug in the upstream catalog lists `input_modalities:
+    // ["text", "image"]` (codex-rs/models-manager/models.json).
+    for (const providerType of [
+      "openai",
+      "admin-openai",
+      "admin-key",
+    ] as const) {
+      for (const model of GPT6_SLUGS) {
+        expect(supportsVisionForConfig(providerType, model), model).toBe(true);
+      }
+    }
+  });
+
   it("rejects non-vision openai models", () => {
     expect(supportsVisionForConfig("openai", "gpt-3.5-turbo")).toBe(false);
     expect(supportsVisionForConfig("openai", "gpt-4")).toBe(false);
@@ -115,6 +137,14 @@ describe("supportsVisionForConfig", () => {
     expect(supportsVisionForConfig("codex", "o3")).toBe(true);
   });
 
+  it("reports vision for codex and admin-codex on the gpt-6 line", () => {
+    for (const providerType of ["codex", "admin-codex"] as const) {
+      for (const model of GPT6_SLUGS) {
+        expect(supportsVisionForConfig(providerType, model), model).toBe(true);
+      }
+    }
+  });
+
   it("mirrors codex vision capability for the shared central codex (admin-codex)", () => {
     // The operator-shared central Codex runs the same wire on the same resolved
     // slug, so its vision gate is identical to `codex`.

src/lib/ai/codex-client.ts

@@ -24,6 +24,12 @@ import {
   invalidateCachedCodexSlug,
   setCachedCodexSlug,
 } from "./codex-slug-cache";
+import {
+  DEFAULT_SLUG_FALLBACK_CHAIN,
+  loadFallbackChain,
+} from "./codex-slug-chain";
+
+export { resolveCodexVisionSlug } from "./codex-slug-chain";
 
 /**
  * Codex backend client — talks to `chatgpt.com/backend-api/codex/responses`
@@ -50,56 +56,6 @@ import {
 
 const CODEX_ENDPOINT = "https://chatgpt.com/backend-api/codex/responses";
 
-/**
- * Default fallback chain, most-current-first as of 2026-10-24. Override via
- * `CODEX_MODEL_FALLBACK_CHAIN` (comma-separated); `CODEX_MODEL` is folded
- * into position 0 when set, duplicates dropped. The backend rotates its
- * ChatGPT-auth slugs without notice: by 2026-10-24 it rejected everything
- * from `gpt-5.5` down and served the `gpt-5.6-*` line. The `gpt-6` line is
- * left out until the reasoning, sampling and vision gates recognise it.
- */
-const DEFAULT_SLUG_FALLBACK_CHAIN = [
-  "gpt-5.6-terra", // current default, verified against the backend 2026-10-24
-  "gpt-5.6-luna", // lighter sibling on the same line
-  "gpt-5.6-sol", // heavier sibling on the same line
-  "gpt-5.5", // previous default, kept as the last rung while plans differ
-] as const;
-
-/**
- * The slug the codex vision gate reasons about *before* a request runs: the
- * cached working slug if one is live, else the first chain entry. Lets the
- * Lab-OCR capability resolver decide whether codex can read images without
- * standing up a client or making a network call. Best-effort — the actual
- * routed slug may differ once the chain walk runs, which the runtime probe +
- * graceful fallback (text/local OCR) covers.
- */
-export function resolveCodexVisionSlug(): string | null {
-  const cached = getCachedCodexSlug();
-  if (cached) return cached;
-  return loadFallbackChain()[0] ?? null;
-}
-
-function loadFallbackChain(): string[] {
-  const envChain = process.env.CODEX_MODEL_FALLBACK_CHAIN?.trim();
-  const fromEnv = envChain
-    ? envChain
-        .split(",")
-        .map((s) => s.trim())
-        .filter(Boolean)
-    : [...DEFAULT_SLUG_FALLBACK_CHAIN];
-  const pinned = process.env.CODEX_MODEL?.trim();
-  const merged = pinned ? [pinned, ...fromEnv] : fromEnv;
-  // Stable de-duplication.
-  const seen = new Set<string>();
-  const result: string[] = [];
-  for (const slug of merged) {
-    if (seen.has(slug)) continue;
-    seen.add(slug);
-    result.push(slug);
-  }
-  return result;
-}
-
 /**
  * Triggers per spec §7b — body / status that mean "this slug is no
  * longer accepted, walk to the next one in the chain".

src/lib/ai/codex-slug-chain.ts

@@ -0,0 +1,69 @@
+/**
+ * The Codex slug ladder: which ChatGPT-auth model slugs the client tries, in
+ * order, and the slug the vision gate reasons about before a request runs.
+ * The walk itself (which statuses move to the next rung, the positive cache)
+ * lives in `codex-client.ts` and `codex-slug-cache.ts`; docs/codex-protocol-spec.md
+ * §7b describes the whole defence.
+ */
+import { getCachedCodexSlug } from "./codex-slug-cache";
+
+/**
+ * Default fallback chain. Override via `CODEX_MODEL_FALLBACK_CHAIN`
+ * (comma-separated); `CODEX_MODEL` is folded into position 0 when set,
+ * duplicates dropped. The backend rotates its ChatGPT-auth slugs without
+ * notice: by 2026-10-24 it rejected everything from `gpt-5.5` down.
+ *
+ * `gpt-5.6-terra` stays first because it is the slug verified against the
+ * backend. The upstream catalog (`codex-rs/models-manager/models.json`)
+ * names `gpt-6-sol` as its upgrade, and the last rotation retired a whole
+ * line at once, so the gpt-6 rungs come next rather than terra's siblings:
+ * `gpt-6-sol` (terra's named successor), `gpt-6.1-sol` (the catalog's
+ * default, also offered on plans `gpt-6-sol` is not), `gpt-6-luna` (the
+ * lighter one). `gpt-6-astra`, the frontier tier, is left out: it spends a
+ * plan's allowance fastest for no gain on these prompts.
+ */
+export const DEFAULT_SLUG_FALLBACK_CHAIN = [
+  "gpt-5.6-terra", // verified against the backend 2026-10-24
+  "gpt-6-sol", // terra's upgrade in the upstream catalog
+  "gpt-6.1-sol", // the catalog default, on every plan
+  "gpt-6-luna", // lighter gpt-6 sibling
+  "gpt-5.6-luna", // lighter sibling on terra's line
+  "gpt-5.6-sol", // heavier sibling on terra's line
+  "gpt-5.5", // previous default, kept as the last rung while plans differ
+] as const;
+
+/**
+ * The slug the codex vision gate reasons about *before* a request runs: the
+ * cached working slug if one is live, else the first chain entry. Lets the
+ * Lab-OCR capability resolver decide whether codex can read images without
+ * standing up a client or making a network call. Best-effort — the actual
+ * routed slug may differ once the chain walk runs, which the runtime probe +
+ * graceful fallback (text/local OCR) covers.
+ */
+export function resolveCodexVisionSlug(): string | null {
+  const cached = getCachedCodexSlug();
+  if (cached) return cached;
+  return loadFallbackChain()[0] ?? null;
+}
+
+/** The chain a client walks: the env override or the default, `CODEX_MODEL` first. */
+export function loadFallbackChain(): string[] {
+  const envChain = process.env.CODEX_MODEL_FALLBACK_CHAIN?.trim();
+  const fromEnv = envChain
+    ? envChain
+        .split(",")
+        .map((s) => s.trim())
+        .filter(Boolean)
+    : [...DEFAULT_SLUG_FALLBACK_CHAIN];
+  const pinned = process.env.CODEX_MODEL?.trim();
+  const merged = pinned ? [pinned, ...fromEnv] : fromEnv;
+  // Stable de-duplication.
+  const seen = new Set<string>();
+  const result: string[] = [];
+  for (const slug of merged) {
+    if (seen.has(slug)) continue;
+    seen.add(slug);
+    result.push(slug);
+  }
+  return result;
+}

src/lib/ai/openai-capabilities.ts

@@ -1,3 +1,5 @@
+import { openAIModelFamily } from "./openai-model-family";
+
 export type OpenAIChatCompletionsCapabilities =
   | {
       tokenBudgetField: "max_completion_tokens";
@@ -18,20 +20,13 @@ const LEGACY_COMPATIBLE_CAPABILITIES = {
   supportsSamplingControls: true,
 } as const;
 
+/**
+ * A reasoning model (gpt-5 on, the gpt-6 line included, the o-series) takes
+ * `max_completion_tokens` and refuses sampling controls; the family reading
+ * is shared with the reasoning and vision gates.
+ */
 function isModernCanonicalModel(model: string): boolean {
-  const normalizedModel = model.toLowerCase();
-
-  return (
-    normalizedModel === "gpt-5" ||
-    normalizedModel.startsWith("gpt-5-") ||
-    normalizedModel.startsWith("gpt-5.") ||
-    normalizedModel === "o1" ||
-    normalizedModel.startsWith("o1-") ||
-    normalizedModel === "o3" ||
-    normalizedModel.startsWith("o3-") ||
-    normalizedModel === "o4" ||
-    normalizedModel.startsWith("o4-")
-  );
+  return openAIModelFamily(model)?.reasoning != null;
 }
 
 export function isCanonicalOpenAIEndpoint(baseUrl: string): boolean {

src/lib/ai/openai-model-family.ts

@@ -0,0 +1,88 @@
+/**
+ * The one reading of an OpenAI model id. Every gate that decides something
+ * from an OpenAI or Codex model name asks this function: the reasoning effort
+ * and its "off" value (`reasoning/support.ts`), the Chat Completions token
+ * field and sampling controls (`openai-capabilities.ts`), and image input
+ * (`vision-capability.ts`). A new generation is taught here once instead of
+ * in three regexes that drift apart.
+ *
+ * The facts behind each branch:
+ *
+ *  - GPT from 5.1 on, the gpt-6 line included, reasons and takes
+ *    `effort: "none"`. A live probe on 2026-10-06 answered `none` on gpt-5.5
+ *    and listed `'none', 'low', 'medium', 'high', 'xhigh'` as supported.
+ *    The upstream Codex catalog (`codex-rs/models-manager/models.json`)
+ *    gives gpt-6-sol, gpt-6-luna, gpt-6-astra and gpt-6.1-sol the same
+ *    reasoning presets as the gpt-5.6 line (`low` to `max`, sol and astra
+ *    also `ultra`). The catalog lists presets, not the wire set: it leaves
+ *    `none` out for gpt-5.5 too, which takes it. A model that does refuse
+ *    `none` is asked again without reasoning and remembers that
+ *    (`reasoning/dialect-cache.ts`), so the assumption costs one retry at
+ *    worst. A later generation (`gpt-7`) is read the same way.
+ *  - Bare gpt-5 (and its mini / nano / dated snapshots) stops at `minimal`.
+ *  - The o-series (o1, o3, o4) starts at `low`.
+ *  - Image input: every GPT from 5 on (the catalog lists
+ *    `input_modalities: ["text", "image"]` for every gpt-5.x and gpt-6
+ *    slug), gpt-4o, gpt-4.1, gpt-4-turbo, and the full o1 / o3 / o4. The
+ *    o-series `-mini` / `-preview` variants are text-only, as are gpt-4 and
+ *    gpt-3.5.
+ *
+ * Client-safe: no server import.
+ */
+
+/**
+ * The reasoning family of an OpenAI model, or `null` for a model that does
+ * not reason. `gpt-5.1+` is every GPT generation from 5.1 on, the gpt-6 line
+ * included.
+ */
+export type OpenAIReasoningFamily = "gpt-5.1+" | "gpt-5" | "o" | null;
+
+export interface OpenAIModelFamily {
+  /** How the model takes a reasoning effort; `null` when it does not reason. */
+  reasoning: OpenAIReasoningFamily;
+  /** The model reads image input. */
+  vision: boolean;
+}
+
+/**
+ * `gpt-<major>[.<minor>][o]`, followed by the end or a `-variant`:
+ * `gpt-6.1-sol`, `gpt-5-codex-mini`, `gpt-4o-mini`, `gpt-4-turbo`.
+ */
+const GPT_ID = /^gpt-(\d+)(?:\.(\d+))?(o?)(?=$|-)/;
+const O_SERIES_ID = /^o[134](?=$|-)/;
+
+/**
+ * What the model family behind `model` can do, or `null` for a name this
+ * file does not recognise as an OpenAI model (a gateway alias, a local
+ * model, a hidden codename slug). An `openai/` gateway prefix is ignored.
+ */
+export function openAIModelFamily(model: string): OpenAIModelFamily | null {
+  const m = model
+    .trim()
+    .toLowerCase()
+    .replace(/^openai\//, "");
+
+  const gpt = GPT_ID.exec(m);
+  if (gpt) {
+    const major = Number(gpt[1]);
+    const minor = gpt[2] === undefined ? null : Number(gpt[2]);
+    const omni = gpt[3] === "o";
+    if (major >= 6 || (major === 5 && minor !== null && minor >= 1)) {
+      return { reasoning: "gpt-5.1+", vision: true };
+    }
+    if (major === 5) return { reasoning: "gpt-5", vision: true };
+    const rest = m.slice(gpt[0].length);
+    const vision =
+      major === 4 && (omni || minor === 1 || rest.startsWith("-turbo"));
+    return { reasoning: null, vision };
+  }
+
+  if (O_SERIES_ID.test(m)) {
+    return {
+      reasoning: "o",
+      vision: !m.includes("-mini") && !m.includes("-preview"),
+    };
+  }
+
+  return null;
+}

src/lib/ai/reasoning/__tests__/support.test.ts

@@ -74,6 +74,11 @@ describe("OpenAI families", () => {
   it.each([
     ["gpt-5.5", "gpt-5.1+", "none"],
     ["gpt-5.1-mini", "gpt-5.1+", "none"],
+    ["gpt-6-sol", "gpt-5.1+", "none"],
+    ["gpt-6-luna", "gpt-5.1+", "none"],
+    ["gpt-6-astra", "gpt-5.1+", "none"],
+    ["gpt-6.1-sol", "gpt-5.1+", "none"],
+    ["openai/gpt-6-sol", "gpt-5.1+", "none"],
     ["gpt-5", "gpt-5", "minimal"],
     ["gpt-5-mini", "gpt-5", "minimal"],
     ["o3", "o", "low"],
@@ -83,6 +88,15 @@ describe("OpenAI families", () => {
     expect(openAIOffEffort(family)).toBe(off);
   });
 
+  it("gives the gpt-6 line an effort and a real off on the operator's key", () => {
+    for (const model of ["gpt-6-sol", "gpt-6.1-sol", "gpt-6-luna"]) {
+      expect(reasoningSupport("admin-key", model), model).toMatchObject({
+        effort: true,
+        offIsReal: true,
+      });
+    }
+  });
+
   it("does not treat a non-reasoning model as one", () => {
     expect(openAIReasoningFamily("gpt-4o")).toBeNull();
     expect(openAIReasoningFamily("gpt-4.1-mini")).toBeNull();

src/lib/ai/reasoning/support.ts

@@ -20,8 +20,8 @@
  *    https://developers.openai.com/cookbook/examples/responses_api/reasoning_items
  *  - OpenAI Chat Completions: `reasoning_effort`, no summaries and no
  *    reasoning items to hand back between tool rounds, so reasoning is only
- *    sent on rounds without tools. `none` exists from gpt-5.1 on; gpt-5 knows
- *    `minimal`; the o-series starts at `low`.
+ *    sent on rounds without tools. `none` exists from gpt-5.1 on (the gpt-6
+ *    line included); gpt-5 knows `minimal`; the o-series starts at `low`.
  *    https://platform.openai.com/docs/api-reference/chat/create
  *  - Anthropic: manual `thinking: { type: "enabled", budget_tokens }` up to
  *    the 4.5 generation; adaptive `thinking: { type: "adaptive" }` with
@@ -46,6 +46,10 @@
  *
  * Client-safe: no server import.
  */
+import {
+  openAIModelFamily,
+  type OpenAIReasoningFamily,
+} from "../openai-model-family";
 import type { ProviderType } from "../types";
 import type { ReasoningLevel } from "./levels";
 
@@ -147,19 +151,17 @@ export function anthropicThinksAlways(model: string): boolean {
 
 // ── OpenAI ──────────────────────────────────────────────────────────────────
 
+export type { OpenAIReasoningFamily };
+
 /**
  * The reasoning family of an OpenAI model id, or `null` for a model that does
- * not reason (`gpt-4o`, `gpt-4.1`). `gpt-5.1+` knows `none`, bare `gpt-5`
- * (and its mini / nano) stops at `minimal`, the o-series at `low`.
+ * not reason (`gpt-4o`, `gpt-4.1`) or is not recognised. `gpt-5.1+` (every
+ * generation from 5.1 on, the gpt-6 line included) knows `none`, bare `gpt-5`
+ * (and its mini / nano) stops at `minimal`, the o-series at `low`. The
+ * reading itself lives in `openai-model-family.ts`.
  */
-export type OpenAIReasoningFamily = "gpt-5.1+" | "gpt-5" | "o" | null;
-
 export function openAIReasoningFamily(model: string): OpenAIReasoningFamily {
-  const m = model.toLowerCase().replace(/^openai\//, "");
-  const gpt5 = /^gpt-5(?:\.(\d+))?(?:$|-)/.exec(m);
-  if (gpt5) return gpt5[1] && Number(gpt5[1]) >= 1 ? "gpt-5.1+" : "gpt-5";
-  if (/^o[134](?:$|-)/.test(m)) return "o";
-  return null;
+  return openAIModelFamily(model)?.reasoning ?? null;
 }
 
 /** The lowest `reasoning_effort` an OpenAI reasoning family accepts. */

src/lib/ai/vision-capability.ts

@@ -12,6 +12,7 @@
  * still resolve to a dead provider at call time (revoked key, unreachable
  * local host); the route surfaces that as an extract failure.
  */
+import { openAIModelFamily } from "./openai-model-family";
 
 /**
  * Anthropic vision models. The Messages API `image` + `document` blocks are
@@ -45,69 +46,22 @@ function anthropicSupportsVision(model: string): boolean {
 }
 
 /**
- * OpenAI vision models (also covers the operator's admin key + Codex when it
- * runs an OpenAI model — though Codex is excluded by `providerType`). `gpt-4o`,
- * `gpt-4.1`, `gpt-4-turbo`, and the full `o1` / `o3` / `o4` reasoning models
- * read image content; `gpt-3.5*` and the legacy `gpt-4` (non-turbo) 400 on an
- * image block.
+ * OpenAI and Codex vision models. Both read the same model families, so both
+ * ask `openAIModelFamily`, which owns the list: every GPT from 5 on (the
+ * gpt-6 line and the `-codex` specialists included), gpt-4o, gpt-4.1,
+ * gpt-4-turbo and the full o1 / o3 / o4. The o-series `-mini` / `-preview`
+ * variants are text-only and must stay excluded, or a user pinned to one is
+ * told OCR works and the provider 400s at the image block; so are gpt-4 and
+ * gpt-3.5.
  *
- * The `o*` reasoning line has text-only variants that DO accept the `o*` prefix
- * but reject an image block: `o1-mini` / `o1-preview` and `o3-mini` have no
- * vision input (only the full `o1` / `o3` are multimodal). They must be
- * excluded or a user pinned to one is told OCR works, then the provider 400s at
- * the image block. We treat an `o1` / `o3` / `o4` slug as vision-capable only
- * when it is NOT a `-mini` / `-preview` variant — the conservative default for
- * any unverified narrow slug.
+ * For Codex (ChatGPT OAuth), image input rides the `codex/responses`
+ * endpoint (docs/codex-protocol-spec.md §2b) with no API key. The slug tested
+ * is the codex client's resolved working slug (cached or chain head), so a
+ * slug rotation onto a model this reading does not know degrades to false
+ * rather than 400-ing the user at the image block.
  */
-function isVisionReasoningSlug(m: string): boolean {
-  if (!(m.startsWith("o1") || m.startsWith("o3") || m.startsWith("o4"))) {
-    return false;
-  }
-  return !m.includes("-mini") && !m.includes("-preview");
-}
-
-function openaiSupportsVision(model: string): boolean {
-  const m = model.toLowerCase();
-  return (
-    m.startsWith("gpt-5") ||
-    m.startsWith("gpt-4o") ||
-    m.startsWith("gpt-4.1") ||
-    m.startsWith("gpt-4-turbo") ||
-    isVisionReasoningSlug(m)
-  );
-}
-
-/**
- * Codex (ChatGPT-OAuth) vision models. Image INPUT is reachable over the
- * `codex/responses` endpoint (docs/codex-protocol-spec.md §2b) on a multimodal
- * slug — no API key needed. The whole GPT-5.x line is text+vision, including
- * the `-codex` specialist slugs (gpt-5.x-codex); only the legacy `gpt-4`-class
- * non-multimodal pins are text-only. The slug we test is the codex client's
- * resolved working slug
- * (cached or chain head), so a slug rotation onto a non-vision model degrades
- * to false rather than 400-ing the user at the image block.
- *
- * `gpt-4o` / `o*` are folded in for completeness (a future codex slug could
- * land there), reusing the same multimodal families the OpenAI gate accepts.
- */
-function codexSupportsVision(model: string): boolean {
-  const m = model.toLowerCase();
-  return (
-    // GPT-5.x line is multimodal (text + vision) — INCLUDING the `-codex`
-    // specialist slugs (gpt-5.x-codex), which gained vision in the 5.2-Codex
-    // line (screenshots / diagrams / charts). Verified against
-    // docs/codex-protocol-spec.md (an `input_image` block on a gpt-5-codex
-    // request) plus OpenAI's current model lineup. The historical
-    // "-codex is text-only" exclusion was stale — the wire reads images on
-    // these slugs, so lab-OCR and document reads both succeed on a `-codex` pin.
-    m.startsWith("gpt-5") ||
-    m.startsWith("gpt-4o") ||
-    m.startsWith("gpt-4.1") ||
-    m.startsWith("gpt-4-turbo") ||
-    // Same `-mini` / `-preview` text-only exclusion as the OpenAI gate: a
-    // future codex slug rotation onto e.g. `o3-mini` must not claim vision.
-    isVisionReasoningSlug(m)
-  );
+function openAIFamilySupportsVision(model: string): boolean {
+  return openAIModelFamily(model)?.vision ?? false;
 }
 
 /**
@@ -130,7 +84,8 @@ export type VisionProviderType =
  * Pure capability check over a (providerType, model) pair.
  *
  *  - anthropic            → vision when the model is a 3.5+/4 family pin.
- *  - openai / admin-*     → vision when the model is a gpt-4o/4.1/turbo/o* pin.
+ *  - openai / admin-*     → vision when the model is a gpt-5+/4o/4.1/turbo/o*
+ *                           pin (the gpt-6 line included).
  *  - local                → trust-by-default true: the operator opted in by
  *                           configuring a self-hosted model and we cannot sniff
  *                           its capabilities; a non-vision local model surfaces
@@ -141,7 +96,7 @@ export type VisionProviderType =
  *                           (and accept incapable ones) on the strength of a
  *                           string the gateway owner picked.
  *  - codex                → vision when the resolved codex slug is a
- *                           multimodal model (GPT-5.x / gpt-4o-class). Image
+ *                           multimodal model (GPT-5.x, gpt-6, gpt-4o-class). Image
  *                           input rides the ChatGPT plan — no API key. A
  *                           non-multimodal slug (or unknown) → false.
  *  - none                 → false (nothing configured).
@@ -156,15 +111,15 @@ export function supportsVisionForConfig(
     case "openai":
     case "admin-openai":
     case "admin-key":
-      return model ? openaiSupportsVision(model) : false;
+      return model ? openAIFamilySupportsVision(model) : false;
     case "local":
     case "openai-compatible":
       return true;
     case "codex":
     // The operator-shared central Codex runs the same `codex/responses` wire on
-    // a resolved GPT-5.x slug, so its vision capability is identical to `codex`.
+    // a resolved Codex slug, so its vision capability is identical to `codex`.
     case "admin-codex":
-      return model ? codexSupportsVision(model) : false;
+      return model ? openAIFamilySupportsVision(model) : false;
     case "none":
     default:
       return false;