@vincemakes/kiso-runtime 0.31.0 → 0.31.1

package/dist/provider/metadata.d.ts

@@ -24,8 +24,15 @@ export interface ModelCapabilities {
     readonly maxOutputTokens: number | null;
     /** "automatic" — the provider caches without request markup (DeepSeek,
      *  OpenAI); "explicit" — the request must place cache_control
-     *  breakpoints (Anthropic); "none" — no caching; null = unknown. */
-    readonly promptCaching: "none" | "automatic" | "explicit" | null;
+     *  breakpoints (Anthropic); "none" — no caching; "unobservable" — the
+     *  endpoint answers a cache figure that carries no information (OR-10,
+     *  owner 2026-09-09: the ChatGPT backend reports `cached_tokens` 0 to
+     *  a third-party client on every request — three identical 1,922-token
+     *  requests, one prompt_cache_key, 0 at every attribution level — so
+     *  its 0 and "not reported" are one value; a display shows no cache
+     *  meter for such a row, while the record keeps what the backend
+     *  said); null = unknown. */
+    readonly promptCaching: "none" | "automatic" | "explicit" | "unobservable" | null;
     /** XP-1: the reasoning capability matrix — supersedes the pre-XP-1
      *  boolean IN PLACE (zero consumers existed, verified). null = unknown:
      *  no mode list, no effort levels, nothing downstream may guess. */
@@ -43,8 +50,10 @@ export interface ModelCapabilities {
  *  some model/effort combinations forbid thinking-disabled. */
 export type ThinkingMode = "default" | "adaptive" | "enabled" | "disabled";
 /** `none` is the OpenAI Responses dialect's zero-reasoning level — a NATIVE
- *  value a request carries, distinct from "default" (no field sent). */
-export type ReasoningEffort = "default" | "none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max";
+ *  value a request carries, distinct from "default" (no field sent).
+ *  `ultra` (OR-6) is the vendor CLI presets' top rung for the ChatGPT
+ *  backend — native THERE only; the first-party model pages stop at `max`. */
+export type ReasoningEffort = "default" | "none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max" | "ultra";
 export interface ReasoningSetting {
     readonly thinking: ThinkingMode;
     readonly effort: ReasoningEffort;

package/dist/provider/metadata.js

@@ -102,42 +102,45 @@ const OPENAI_MODELS_ASOF = "2026-09-08";
  *  subscription; a first-party model page says nothing about that backend). */
 const CHATGPT_PRESETS_SOURCE = "https://github.com/openai/codex/blob/37f4bb94c9f4e180535a12e3f2c2f93f4a773df0/codex-rs/models-manager/models.json";
 const RESPONSES_WIRE = "reasoning.effort";
-/** A first-party Responses row: the model page is the source for the
- *  levels, the window, the output cap and the price alike (read the same
- *  day). The adapter asks for no reasoning summaries (encrypted items
- *  only), so no thinking stream is emitted. */
-const openaiRow = (model, effortDefault, rate, page) => ({
+const openaiRow = (model, effortDefault, rate, page, opts = {}) => ({
     model,
     providerId: "openai",
     endpoint: "https://api.openai.com",
     capabilities: { contextWindow: 1_050_000, maxOutputTokens: 128_000, promptCaching: "automatic", reasoning: {
             emitsThinkingStream: false,
             thinking: null,
-            effort: { levels: ["none", "low", "medium", "high", "xhigh"], default: effortDefault, wire: RESPONSES_WIRE },
-            asOf: OPENAI_MODELS_ASOF,
+            effort: { levels: opts.levels ?? ["none", "low", "medium", "high", "xhigh"], default: effortDefault, wire: RESPONSES_WIRE },
+            asOf: opts.asOf ?? OPENAI_MODELS_ASOF,
             source: page,
         }, inputModalities: null },
-    capabilitiesAsOf: OPENAI_MODELS_ASOF,
+    capabilitiesAsOf: opts.asOf ?? OPENAI_MODELS_ASOF,
     capabilitiesSource: page,
-    pricing: { ...rate, cacheWritePerM: 0, asOf: OPENAI_MODELS_ASOF, source: page },
+    pricing: { ...rate, cacheWritePerM: 0, asOf: opts.asOf ?? OPENAI_MODELS_ASOF, source: page },
 });
 /** The subscription backend's row for the same id: the presets' four
- *  levels (no `none`), the presets' context window, no price. */
-const chatgptRow = (model) => ({
+ *  levels (no `none`), the presets' context window, no price, and a cache
+ *  figure no client can observe (OR-10 — see ModelCapabilities). */
+const chatgptRow = (model, opts = {}) => ({
     model,
     providerId: "chatgpt",
     endpoint: "https://chatgpt.com",
-    capabilities: { contextWindow: 272_000, maxOutputTokens: null, promptCaching: null, reasoning: {
+    capabilities: { contextWindow: 272_000, maxOutputTokens: null, promptCaching: "unobservable", reasoning: {
             emitsThinkingStream: false,
             thinking: null,
-            effort: { levels: ["low", "medium", "high", "xhigh"], default: "medium", wire: RESPONSES_WIRE },
-            asOf: OPENAI_MODELS_ASOF,
-            source: CHATGPT_PRESETS_SOURCE,
+            effort: { levels: opts.levels ?? ["low", "medium", "high", "xhigh"], default: opts.default ?? "medium", wire: RESPONSES_WIRE },
+            asOf: opts.asOf ?? OPENAI_MODELS_ASOF,
+            source: opts.source ?? CHATGPT_PRESETS_SOURCE,
         }, inputModalities: null },
-    capabilitiesAsOf: OPENAI_MODELS_ASOF,
-    capabilitiesSource: CHATGPT_PRESETS_SOURCE,
+    capabilitiesAsOf: opts.asOf ?? OPENAI_MODELS_ASOF,
+    capabilitiesSource: opts.source ?? CHATGPT_PRESETS_SOURCE,
     pricing: null,
 });
+/** OR-6 (2026-09-09): the newer line, read the same day — the model pages for
+ *  the first-party rows, the presets pinned to a later commit for the
+ *  subscription rows (both ids answered on the backend that day). */
+const OPENAI_MODELS_ASOF_2 = "2026-09-09";
+const CHATGPT_PRESETS_SOURCE_2 = "https://github.com/openai/codex/blob/634ebc1865c6ac840ed3ba118f040d527bf4b55d/codex-rs/models-manager/models.json";
+const ULTRA_LADDER = ["low", "medium", "high", "xhigh", "max", "ultra"];
 /** The v1 table. Nulls outnumber numbers ON PURPOSE: only values with a
  *  named source enter; everything else waits for one. */
 const ENTRIES = [
@@ -232,6 +235,17 @@ const ENTRIES = [
     openaiRow("gpt-5.4", "none", { inputPerM: 2.5, outputPerM: 15, cacheReadPerM: 0.25 }, "https://developers.openai.com/api/docs/models/gpt-5.4"),
     chatgptRow("gpt-5.5"),
     chatgptRow("gpt-5.4"),
+    // OR-6 (2026-09-09): gpt-6-astra and gpt-5.6-sol. The astra page names
+    // no default, so the first-party default stays null (unknown is null);
+    // the presets say `low` at the subscription. `ultra` exists only there.
+    // The ordering keeps the first-party row first for endpoint-less
+    // callers; the run-side resolver passes the endpoint (45a9df5), so the
+    // two rows need not be nested — astra's subscription row has a level
+    // (ultra) its first-party row lacks.
+    openaiRow("gpt-6-astra", null, { inputPerM: 10, outputPerM: 50, cacheReadPerM: 1 }, "https://developers.openai.com/api/docs/models/gpt-6-astra", { levels: ["low", "medium", "high", "xhigh", "max"], asOf: OPENAI_MODELS_ASOF_2 }),
+    openaiRow("gpt-5.6-sol", "medium", { inputPerM: 4, outputPerM: 20, cacheReadPerM: 0.4 }, "https://developers.openai.com/api/docs/models/gpt-5.6-sol", { levels: ["none", "low", "medium", "high", "xhigh", "max"], asOf: OPENAI_MODELS_ASOF_2 }),
+    chatgptRow("gpt-6-astra", { levels: ULTRA_LADDER, default: "low", source: CHATGPT_PRESETS_SOURCE_2, asOf: OPENAI_MODELS_ASOF_2 }),
+    chatgptRow("gpt-5.6-sol", { levels: ULTRA_LADDER, default: "low", source: CHATGPT_PRESETS_SOURCE_2, asOf: OPENAI_MODELS_ASOF_2 }),
 ];
 /**
  * Look a model up. `endpoint` narrows: an entry WITH an endpoint only

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vincemakes/kiso-runtime",
-	"version": "0.31.0",
+	"version": "0.31.1",
 	"description": "kiso runtime — durable multi-turn agent sessions: AgentDefinition, AgentRuntime, AgentSession, Run, append-only JSONL store.",
 	"type": "module",
 	"license": "MIT",
@@ -25,12 +25,12 @@
 		"test": "vitest run"
 	},
 	"dependencies": {
-		"@vincemakes/kiso-core": "0.31.0"
+		"@vincemakes/kiso-core": "0.31.1"
 	},
 	"peerDependencies": {
-		"@vincemakes/kiso-provider-anthropic": "0.31.0",
-		"@vincemakes/kiso-provider-openai": "0.31.0",
-		"@vincemakes/kiso-provider-openai-responses": "0.31.0"
+		"@vincemakes/kiso-provider-anthropic": "0.31.1",
+		"@vincemakes/kiso-provider-openai": "0.31.1",
+		"@vincemakes/kiso-provider-openai-responses": "0.31.1"
 	},
 	"peerDependenciesMeta": {
 		"@vincemakes/kiso-provider-anthropic": {
@@ -44,7 +44,7 @@
 		}
 	},
 	"devDependencies": {
-		"@vincemakes/kiso-evals": "0.31.0",
+		"@vincemakes/kiso-evals": "0.31.1",
 		"@types/node": "^26.1.2",
 		"typescript": "^5.7.2",
 		"vitest": "^3.0.0"