Merge branch 'main' into posthog-code/fix-archive-story-timestamps

Author: richardsolomou

packages/agent/src/adapters/base-acp-agent.ts

@@ -18,11 +18,11 @@ import type {
 } from "@agentclientprotocol/sdk";
 import { restrictedModelMeta } from "@posthog/shared";
 import {
+  compareModelsForPicker,
   DEFAULT_GATEWAY_MODEL,
   fetchGatewayModels,
   formatGatewayModelName,
   type GatewayModel,
-  getClaudeModelRecency,
   isAnthropicModel,
   isCloudflareModel,
   isCloudflareModelId,
@@ -163,12 +163,7 @@ export abstract class BaseAcpAgent implements Agent {
         // silently dropping them.
         ...(model.allowed ? {} : { _meta: restrictedModelMeta() }),
       }))
-      // Sort oldest-to-newest so the picker is deterministic and the newest
-      // model lands at the end of the list, closest to the trigger.
-      .sort(
-        (a, b) =>
-          getClaudeModelRecency(a.value) - getClaudeModelRecency(b.value),
-      );
+      .sort((a, b) => compareModelsForPicker(a.value, b.value));
 
     // Models the Claude adapter can drive: Anthropic ids, plus Cloudflare `@cf/` ids the gateway
     // serves over its Anthropic-Messages surface. Anything else (e.g. a Codex/GPT id) is a genuine

packages/agent/src/adapters/claude/session/model-config.test.ts

@@ -19,6 +19,18 @@ describe("applyAvailableModelsAllowlist", () => {
     ).toEqual(rawModelOptions);
   });
 
+  it("reorders the allowlist to the picker order instead of the listed order", () => {
+    expect(
+      applyAvailableModelsAllowlist(rawModelOptions, [
+        "claude-sonnet-4-6",
+        "claude-opus-4-8",
+      ]).options,
+    ).toEqual([
+      { value: "claude-opus-4-8", name: "Claude Opus 4.8" },
+      { value: "claude-sonnet-4-6", name: "Claude Sonnet 4.6" },
+    ]);
+  });
+
   it("switches the current model when the previous one is filtered out", () => {
     expect(
       applyAvailableModelsAllowlist(rawModelOptions, ["claude-sonnet-4-6"]),

packages/agent/src/adapters/claude/session/model-config.ts

@@ -1,4 +1,5 @@
 import type { SessionConfigSelectOption } from "@agentclientprotocol/sdk";
+import { compareModelsForPicker } from "../../../gateway-models";
 
 export interface ModelConfigOptions {
   currentModelId: string;
@@ -30,6 +31,9 @@ export function applyAvailableModelsAllowlist(
 
   if (filtered.length === 0) return modelOptions;
 
+  // The allowlist is a filter, not a display order: keep the picker's ordering.
+  filtered.sort((a, b) => compareModelsForPicker(a.value, b.value));
+
   const currentModelId = filtered.some(
     (o) => o.value === modelOptions.currentModelId,
   )

packages/agent/src/gateway-models.test.ts

@@ -1,5 +1,6 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
 import {
+  compareModelsForPicker,
   fetchGatewayModels,
   fetchModelsList,
   formatGatewayModelName,
@@ -34,7 +35,7 @@ describe("formatGatewayModelName", () => {
     ).toBe("Claude Opus 4.8");
   });
 
-  it("formats OpenAI models as raw lowercase model ids", () => {
+  it("uppercases the GPT acronym in OpenAI model ids", () => {
     expect(
       formatGatewayModelName({
         id: "GPT-5.5",
@@ -44,23 +45,23 @@ describe("formatGatewayModelName", () => {
         supports_vision: true,
         allowed: true,
       }),
-    ).toBe("gpt-5.5");
+    ).toBe("GPT-5.5");
   });
 
-  it("strips the openai/ prefix from OpenAI model ids", () => {
+  it("strips the openai/ prefix, uppercases GPT, and title-cases the suffix", () => {
     expect(
       formatGatewayModelName({
-        id: "openai/gpt-5.5",
+        id: "openai/gpt-5.6-sol",
         owned_by: "openai",
         context_window: 200000,
         supports_streaming: true,
         supports_vision: true,
         allowed: true,
       }),
-    ).toBe("gpt-5.5");
+    ).toBe("GPT-5.6 Sol");
   });
 
-  it("formats Cloudflare models as the lowercase final path segment", () => {
+  it("formats Cloudflare models as the final path segment with GLM uppercased", () => {
     expect(
       formatGatewayModelName({
         id: "@cf/zai-org/glm-5.2",
@@ -70,7 +71,20 @@ describe("formatGatewayModelName", () => {
         supports_vision: false,
         allowed: true,
       }),
-    ).toBe("glm-5.2");
+    ).toBe("GLM-5.2");
+  });
+
+  it("leaves non-acronym Cloudflare models lowercase", () => {
+    expect(
+      formatGatewayModelName({
+        id: "@cf/meta/llama-3.1-8b-instruct",
+        owned_by: "cloudflare",
+        context_window: 128000,
+        supports_streaming: true,
+        supports_vision: false,
+        allowed: true,
+      }),
+    ).toBe("llama-3.1-8b-instruct");
   });
 
   it("blocks deprecated Claude gateway models", () => {
@@ -115,28 +129,28 @@ describe("getClaudeModelRecency", () => {
       getClaudeModelRecency("claude-fable-5"),
     );
   });
+});
 
-  it("produces the full picker display order, oldest to newest", () => {
+describe("compareModelsForPicker", () => {
+  it("groups models by family, most capable first, newest version first", () => {
     // Models as the gateway might return them — arbitrary order.
     const gatewayOrder = [
       "claude-fable-5",
-      "claude-opus-4-8",
+      "claude-opus-4-7",
       "claude-mystery",
+      "claude-sonnet-5",
       "claude-haiku-4-5",
       "claude-sonnet-4-6",
-      "claude-opus-4-7",
+      "claude-opus-4-8",
     ];
-    const displayed = [...gatewayOrder].sort(
-      (a, b) => getClaudeModelRecency(a) - getClaudeModelRecency(b),
-    );
-    // The menu opens upward, so the newest model (last here) sits closest to
-    // the trigger. Unknown/unversioned models rank newest and trail the list.
+    const displayed = [...gatewayOrder].sort(compareModelsForPicker);
     expect(displayed).toEqual([
-      "claude-haiku-4-5",
-      "claude-sonnet-4-6",
-      "claude-opus-4-7",
-      "claude-opus-4-8",
       "claude-fable-5",
+      "claude-opus-4-8",
+      "claude-opus-4-7",
+      "claude-sonnet-5",
+      "claude-sonnet-4-6",
+      "claude-haiku-4-5",
       "claude-mystery",
     ]);
   });

packages/agent/src/gateway-models.ts

@@ -258,16 +258,8 @@ export function getProviderName(ownedBy: string): string {
   return PROVIDER_NAMES[ownedBy] ?? ownedBy;
 }
 
-// Sort key for ordering models oldest-to-newest in pickers. The model menu
-// opens upward (side="top"), so the last item sits closest to the trigger —
-// sorting ascending by this key puts the newest model right under the user's
-// cursor. The key is the version embedded in the model id, e.g.
-// "claude-sonnet-4-6" -> 4006, "claude-opus-4-8" -> 4008, "claude-fable-5" ->
-// 5000; a higher number means a newer model. An id with no recognisable
-// version (a brand-new or unexpected release) ranks as newest so it still
-// surfaces at the end rather than at an arbitrary gateway-determined position.
-// Only the first version group is read, so a trailing date suffix (e.g.
-// "-20251001") is ignored; the minor component is assumed to be < 1000.
+// Version embedded in the model id, e.g. "claude-opus-4-8" -> 4008. Ids with no
+// recognisable version rank newest. A trailing date suffix is ignored.
 export function getClaudeModelRecency(modelId: string): number {
   const match = modelId.toLowerCase().match(/-(\d+)(?:[-.](\d+))?/);
   if (!match) return Number.MAX_SAFE_INTEGER;
@@ -276,15 +268,48 @@ export function getClaudeModelRecency(modelId: string): number {
   return major * 1000 + minor;
 }
 
+// Families ordered most-capable first; unknown families sort last.
+const MODEL_FAMILY_ORDER = ["fable", "opus", "sonnet", "haiku"];
+
+function getModelFamilyRank(modelId: string): number {
+  const id = modelId.toLowerCase();
+  const index = MODEL_FAMILY_ORDER.findIndex((family) => id.includes(family));
+  return index === -1 ? MODEL_FAMILY_ORDER.length : index;
+}
+
+// Group by family, then newest version first within each family.
+export function compareModelsForPicker(a: string, b: string): number {
+  const familyDiff = getModelFamilyRank(a) - getModelFamilyRank(b);
+  if (familyDiff !== 0) return familyDiff;
+  return getClaudeModelRecency(b) - getClaudeModelRecency(a);
+}
+
 const PROVIDER_PREFIXES = ["anthropic/", "openai/", "google-vertex/"];
 
+const KNOWN_ACRONYMS = new Set(["gpt", "glm"]);
+
+// For a known acronym, uppercase it, keep the version attached, and title-case
+// any suffix: "gpt-5.6-sol" -> "GPT-5.6 Sol", "glm-5.2" -> "GLM-5.2". Other ids
+// stay lowercase to avoid mangling ordinary names (e.g. "llama-3.1-8b").
+function formatProviderModelName(modelId: string): string {
+  const [acronym, version, ...suffix] = modelId.split("-");
+  if (!KNOWN_ACRONYMS.has(acronym.toLowerCase())) return modelId.toLowerCase();
+  const head = version
+    ? `${acronym.toUpperCase()}-${version}`
+    : acronym.toUpperCase();
+  const tail = suffix.map(
+    (word) => word.charAt(0).toUpperCase() + word.slice(1).toLowerCase(),
+  );
+  return [head, ...tail].join(" ");
+}
+
 export function formatGatewayModelName(model: GatewayModel): string {
   if (isCloudflareModel(model)) {
-    return (model.id.split("/").pop() ?? model.id).toLowerCase();
+    return formatProviderModelName(model.id.split("/").pop() ?? model.id);
   }
 
   if (isOpenAIModel(model)) {
-    return stripProviderPrefix(model.id).toLowerCase();
+    return formatProviderModelName(stripProviderPrefix(model.id));
   }
 
   return formatModelId(model.id);