fix(models): use the 1M context window Qoder actually supports (#13)

auto, efficient and lite carried a 180K contextWindow, so pi showed >100%
context usage and auto-compacted early. Live testing in #13 drove global
lite to 1,000K tokens successfully, so the 180K value was a stale
max_input_tokens floor rather than a real limit.

Introduce DEFAULT_CONTEXT_WINDOW (1M) and use it for the global static
models, including gm51model. Apply it as the catalog fallback too, but only
when /model/list advertises no context_config: models that do advertise a
smaller window (kmodel at 256K, CN GLM-5.2 and MiniMax at 200K) keep it
rather than being inflated to 1M. CN Auto stays at 200K since only global
lite was verified at 1M.

Co-authored-by: Simon Shi <simonsmh@users.noreply.github.com>
This commit is contained in:
Cursor Agent
2026-08-31 15:57:49 +00:00
co-authored by Simon Shi
parent 1dd1b0988b
commit 1e44cde685
4 changed files with 159 additions and 36 deletions
+43
View File
@@ -71,4 +71,47 @@ describe("Qoder model cache", () => {
expect(getCachedModels("global").map((model) => model.id)).toEqual(["ultimate"]);
});
it("records a 1M context window when the catalog omits context_config", async () => {
vi.stubGlobal(
"fetch",
vi.fn().mockResolvedValue({
ok: true,
json: () =>
Promise.resolve({
chat: [{ key: "lite", enable: true, display_name: "Lite", max_input_tokens: 180000 }],
}),
}),
);
await updateQoderModelsCache("access-token", "user-id", "Test User", "test@example.com", "global");
const cache = JSON.parse(readFileSync(CACHE_PATH, "utf8"));
expect(cache.models[0].contextWindow).toBe(1_000_000);
});
it("records the advertised context_config max, even when it is below 1M", async () => {
vi.stubGlobal(
"fetch",
vi.fn().mockResolvedValue({
ok: true,
json: () =>
Promise.resolve({
chat: [
{
key: "gm51model",
enable: true,
display_name: "GLM 5.2",
context_config: { default: { token_count: 200000, is_default: true } },
},
],
}),
}),
);
await updateQoderModelsCache("access-token", "user-id", "Test User", "test@example.com", "global");
const cache = JSON.parse(readFileSync(CACHE_PATH, "utf8"));
expect(cache.models[0].contextWindow).toBe(200000);
});
});
+54 -1
View File
@@ -1,5 +1,11 @@
import { describe, expect, it } from "vitest";
import { staticCnModels, staticModels, ZERO_COST } from "../models.js";
import {
contextWindowFromCatalog,
DEFAULT_CONTEXT_WINDOW,
staticCnModels,
staticModels,
ZERO_COST,
} from "../models.js";
// ── staticModels ──────────────────────────────────────────────────────────
@@ -33,6 +39,21 @@ describe("staticModels", () => {
const ids = staticModels.map((m) => m.id);
expect(new Set(ids).size).toBe(ids.length);
});
it("uses a 1M context window for models confirmed to support it", () => {
// Global lite was live-tested through 1,000K tokens (issue #13). auto,
// efficient, and gm51model share the same Qoder 1M catalog family.
for (const id of ["auto", "efficient", "lite", "gm51model"]) {
const model = staticModels.find((m) => m.id === id);
expect(model, id).toBeDefined();
expect(model?.contextWindow).toBe(DEFAULT_CONTEXT_WINDOW);
expect(model?.contextWindow).toBe(1_000_000);
}
});
it("keeps kmodel at the catalog-advertised 256K window", () => {
expect(staticModels.find((m) => m.id === "kmodel")?.contextWindow).toBe(256000);
});
});
// ── staticCnModels ────────────────────────────────────────────────────────
@@ -73,6 +94,38 @@ describe("staticCnModels", () => {
expect(m.description).toBeTruthy();
}
});
it("does not copy global 1M onto CN models whose live catalog is smaller", () => {
expect(staticCnModels.find((m) => m.id === "Auto")?.contextWindow).toBe(200000);
expect(staticCnModels.find((m) => m.id === "GLM-5.2")?.contextWindow).toBe(200000);
expect(staticCnModels.find((m) => m.id === "MiniMax-M2.7")?.contextWindow).toBe(200000);
expect(staticCnModels.find((m) => m.id === "Kimi-K2.7-Code")?.contextWindow).toBe(256000);
});
});
describe("contextWindowFromCatalog", () => {
it("uses the largest advertised context_config token_count", () => {
expect(
contextWindowFromCatalog({
context_config: {
small: { token_count: 200000 },
large: { token_count: 1000000, is_default: true },
},
}),
).toBe(1000000);
});
it("keeps an advertised 200K window instead of the 1M fallback", () => {
expect(
contextWindowFromCatalog({
context_config: { default: { token_count: 200000, is_default: true } },
}),
).toBe(200000);
});
it("falls back to 1M when the catalog omits context_config", () => {
expect(contextWindowFromCatalog({ key: "lite", max_input_tokens: 180000 })).toBe(DEFAULT_CONTEXT_WINDOW);
});
});
// ── ZERO_COST ─────────────────────────────────────────────────────────────
+56 -34
View File
@@ -26,6 +26,16 @@ export const ZERO_COST = Object.freeze({ input: 0, output: 0, cacheRead: 0, cach
*/
export const MAX_OUTPUT_TOKENS = 131072;
/**
* Fallback context window when the catalog omits `context_config`.
*
* Qoder's `/model/list` often ships `max_input_tokens` as a stale 180K floor
* even for models that accept 1M-token prompts (verified against global `lite`
* through 1,000K tokens). When `context_config` is present we use its largest
* `token_count` instead, so models that truly advertise 200K/256K stay there.
*/
export const DEFAULT_CONTEXT_WINDOW = 1000000;
/** Shape of a single entry returned by the Qoder /model/list endpoint. */
export interface QoderModelEntry {
key?: string;
@@ -79,7 +89,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 180000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -92,7 +102,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: true,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -105,7 +115,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: true,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -118,7 +128,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 180000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -131,7 +141,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text"],
cost: ZERO_COST,
contextWindow: 180000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -144,7 +154,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -157,7 +167,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: true,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -170,7 +180,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: true,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -183,7 +193,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -196,7 +206,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: true,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -209,7 +219,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: true,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -222,7 +232,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: true,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -235,6 +245,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
// Catalog advertises 256K; not included in the 1M live test in issue #13.
contextWindow: 256000,
maxTokens: MAX_OUTPUT_TOKENS,
},
@@ -248,7 +259,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
{
@@ -261,7 +272,7 @@ export const staticModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
},
];
@@ -277,6 +288,8 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
// CN Auto has not been live-tested at 1M; keep the conservative 200K
// fallback until the CN catalog advertises a larger option.
contextWindow: 200000,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN smart routing; fallback context window of 200K.",
@@ -291,7 +304,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN qmodel_latest; context options 200K/400K/1M.",
},
@@ -305,7 +318,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN qmodel; context options 200K/400K/1M.",
},
@@ -319,7 +332,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN q36fmodel; context options 200K/400K/1M.",
},
@@ -333,7 +346,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN dmodel; context options 200K/400K/1M.",
},
@@ -347,7 +360,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text"],
cost: ZERO_COST,
contextWindow: 1000000,
contextWindow: DEFAULT_CONTEXT_WINDOW,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN dfmodel; context options 200K/400K/1M.",
},
@@ -361,6 +374,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
// Live CN catalog currently displays 200K; do not copy global gm51model's 1M.
contextWindow: 200000,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN gm51model; live catalog currently displays GLM-5.2 with 200K context.",
@@ -375,6 +389,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text", "image"],
cost: ZERO_COST,
// Catalog advertises 256K; same as global kmodel.
contextWindow: 256000,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN kmodel; context option 256K.",
@@ -389,6 +404,7 @@ export const staticCnModels: QoderModelDef[] = [
supportsEffort: false,
input: ["text"],
cost: ZERO_COST,
// Live CN catalog reports 200K; not confirmed at 1M.
contextWindow: 200000,
maxTokens: MAX_OUTPUT_TOKENS,
description: "Qoder CN mmodel; live catalog reports 200K context.",
@@ -485,6 +501,21 @@ export function getCachedModelConfig(modelKey: string, mode?: string): QoderMode
return null;
}
/** Resolve contextWindow from a catalog entry. Exported for tests. */
export function contextWindowFromCatalog(entry: QoderModelEntry): number {
const contextConfig = entry.context_config;
if (contextConfig && typeof contextConfig === "object") {
let advertised = 0;
for (const configVal of Object.values(contextConfig)) {
if (configVal && typeof configVal === "object" && typeof configVal.token_count === "number") {
if (configVal.token_count > advertised) advertised = configVal.token_count;
}
}
if (advertised > 0) return advertised;
}
return DEFAULT_CONTEXT_WINDOW;
}
/** Prefer the largest context option when Qoder exposes selectable contexts. */
function withMaxContextAsDefault(entry: QoderModelEntry): QoderModelEntry {
const contextConfig = entry.context_config;
@@ -559,21 +590,12 @@ export async function updateQoderModelsCache(
if (!key || !entry.enable) continue;
const display = entry.display_name || key;
// The context window is the largest selectable context option the upstream
// catalog exposes (e.g. 1M when 200K/400K/1M are offered). Fall back to
// 200K when no context_config is present. `max_input_tokens` is not used:
// it is a flat base value (180K) that every catalog entry with selectable
// contexts already exceeds, so it only ever acted as a redundant floor.
let ctxLen = 200000;
if (entry.context_config && typeof entry.context_config === "object") {
for (const configVal of Object.values(entry.context_config)) {
if (configVal && typeof configVal === "object" && typeof configVal.token_count === "number") {
if (configVal.token_count > ctxLen) {
ctxLen = configVal.token_count;
}
}
}
}
// Prefer the largest selectable context option the catalog advertises
// (e.g. 1M when 200K/400K/1M are offered). If none is advertised, use
// DEFAULT_CONTEXT_WINDOW rather than the stale 180K `max_input_tokens`
// floor. Do not seed from DEFAULT_CONTEXT_WINDOW before scanning
// context_config: that would inflate models that only advertise 200K.
const ctxLen = contextWindowFromCatalog(entry);
const isVL = !!entry.is_vl;
const isReasoning = !!entry.is_reasoning || !!entry.thinking_config;
const supportsEffort = !!entry.thinking_config?.enabled?.efforts;