From 1e44cde685173f3d3f68805e99770075a6e718c9 Mon Sep 17 00:00:00 2001 From: Cursor Agent Date: Mon, 31 Aug 2026 15:57:49 +0000 Subject: [PATCH] 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 --- README.md | 7 ++- src/__tests__/models-cache.test.ts | 43 ++++++++++++++ src/__tests__/models.test.ts | 55 +++++++++++++++++- src/models.ts | 90 +++++++++++++++++++----------- 4 files changed, 159 insertions(+), 36 deletions(-) diff --git a/README.md b/README.md index 9f62e7f..9e149e0 100644 --- a/README.md +++ b/README.md @@ -111,6 +111,11 @@ Exposes the backing model keys returned by Qoder, including: - `kmodel_latest` (Kimi K3) - `mmodel` (MiniMax M3) +Global models default to a 1M context window, matching the Qoder API (verified +against `lite` with prompts up to 1,000K tokens). `kmodel` stays at the 256K its +catalog entry advertises. Once you log in, the live `/model/list` catalog +overrides these fallbacks with the largest context option each model exposes. + ### China `qoder-cn` The China provider exposes friendly model IDs and maps them back to Qoder CN's @@ -118,7 +123,7 @@ internal keys at request time: | Friendly ID | Qoder CN key | Context | Images | Reasoning | | --- | --- | ---: | :---: | :---: | -| `auto` | `auto` | 180K | ✅ | ✅ | +| `auto` | `auto` | 200K | ✅ | ✅ | | `qwen3.7-max` | `qmodel_latest` | 1M | ✅ | ✅ | | `qwen3.7-plus` | `qmodel` | 1M | ❌ | ✅ | | `qwen3.6-flash` | `q36fmodel` | 1M | ❌ | ✅ | diff --git a/src/__tests__/models-cache.test.ts b/src/__tests__/models-cache.test.ts index 793daa0..168edc4 100644 --- a/src/__tests__/models-cache.test.ts +++ b/src/__tests__/models-cache.test.ts @@ -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); + }); }); diff --git a/src/__tests__/models.test.ts b/src/__tests__/models.test.ts index 167c93e..4caa98c 100644 --- a/src/__tests__/models.test.ts +++ b/src/__tests__/models.test.ts @@ -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 ───────────────────────────────────────────────────────────── diff --git a/src/models.ts b/src/models.ts index f13019b..339641f 100644 --- a/src/models.ts +++ b/src/models.ts @@ -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;