Skip to content

Never stop coding. Free MIT AI gateway: one endpoint, 375 providers (150+ free), 1,400+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 900+ contributors

fix(catalog): auto-calculate combo context_length from target model limits

Automation3dc7542eca40
diff --git a/src/app/api/v1/models/catalog.ts b/src/app/api/v1/models/catalog.ts
index 12fe5b302c..747253635f 100644
--- a/src/app/api/v1/models/catalog.ts
+++ b/src/app/api/v1/models/catalog.ts
@@ -25,6 +25,8 @@ import {
getCatalogDiagnosticsHeaders,
} from "@/lib/modelMetadataRegistry";
import { isAuthRequired, isDashboardSessionAuthenticated } from "@/shared/utils/apiAuth";
+import { parseModel } from "@omniroute/open-sse/services/model.ts";
+import { getTokenLimit } from "@omniroute/open-sse/services/contextManager.ts";
const FALLBACK_ALIAS_TO_PROVIDER = {
ag: "antigravity",
@@ -313,6 +315,30 @@ export async function getUnifiedModelsResponse(
// Add combos first (they appear at the top) — only active ones
for (const combo of combos) {
if (combo.isActive === false || combo.isHidden === true) continue;
+
+ // Calculate combo context length from its model targets.
+ // OpenCode and other clients read context_length from the catalog; without it
+ // they fall back to a conservative ~4000 token limit, causing truncation.
+ const comboContextLength = Array.isArray(combo.models)
+ ? combo.models
+ .filter((step) => step && step.kind === "model" && step.model)
+ .map((step) => {
+ const parsed = parseModel(step.model);
+ const provider = parsed.provider || (step as any).providerId || "unknown";
+ const model = parsed.model || step.model;
+ return getTokenLimit(provider, model);
+ })
+ .filter((limit): limit is number => typeof limit === "number" && limit > 0)
+ .reduce((min, limit) => Math.min(min, limit), Infinity)
+ : undefined;
+
+ const effectiveContextLength =
+ typeof combo.context_length === "number" && combo.context_length > 0
+ ? combo.context_length
+ : comboContextLength !== undefined && comboContextLength !== Infinity
+ ? comboContextLength
+ : undefined;
+
models.push({
id: combo.name,
object: "model",
@@ -321,7 +347,7 @@ export async function getUnifiedModelsResponse(
permission: [],
root: combo.name,
parent: null,
- ...(combo.context_length ? { context_length: combo.context_length } : {}),
+ ...(effectiveContextLength !== undefined ? { context_length: effectiveContextLength } : {}),
});
}
diff --git a/tests/unit/models-catalog-route.test.ts b/tests/unit/models-catalog-route.test.ts
index 1ca0db09c8..d0036c91b9 100644
--- a/tests/unit/models-catalog-route.test.ts
+++ b/tests/unit/models-catalog-route.test.ts
@@ -892,3 +892,57 @@ test("v1 models catalog adds managed fallback models for Claude-compatible provi
assert.ok(ids.has("ccdemo/claude-opus-4-6"));
assert.equal(ids.has("ccdemo/claude-sonnet-4-6"), false);
});
+
+test("v1 models catalog auto-calculates combo context_length from targets when not set manually", async () => {
+ await seedConnection("openai", { name: "openai-auto-context" });
+ await seedConnection("claude", {
+ authType: "oauth",
+ name: "claude-auto-context",
+ apiKey: null,
+ accessToken: "claude-access",
+ });
+
+ // Create a combo with targets having different context limits.
+ // openai/gpt-4o context = 128000, claude/claude-sonnet-4-6 = 200000.
+ // The combo should expose context_length = min = 128000.
+ const combo = await combosDb.createCombo({
+ name: "auto-context-combo",
+ strategy: "priority",
+ models: ["openai/gpt-4o", "claude/claude-sonnet-4-6"],
+ });
+
+ const response = await v1ModelsCatalog.getUnifiedModelsResponse(
+ new Request("http://localhost/api/v1/models")
+ );
+ const body = (await response.json()) as any;
+ const comboModel = body.data.find((item) => item.id === "auto-context-combo");
+
+ assert.equal(response.status, 200);
+ assert.ok(comboModel);
+ assert.equal(
+ comboModel.context_length,
+ 128000,
+ "combo context_length should be the MIN of all target model limits"
+ );
+});
+
+test("v1 models catalog prefers manual combo context_length over auto-calculated", async () => {
+ await seedConnection("openai", { name: "openai-manual-context" });
+
+ const combo = await combosDb.createCombo({
+ name: "manual-context-combo",
+ strategy: "priority",
+ models: ["openai/gpt-4o"],
+ });
+ await combosDb.updateCombo((combo as any).id, { context_length: 64000 });
+
+ const response = await v1ModelsCatalog.getUnifiedModelsResponse(
+ new Request("http://localhost/api/v1/models")
+ );
+ const body = (await response.json()) as any;
+ const comboModel = body.data.find((item) => item.id === "manual-context-combo");
+
+ assert.equal(response.status, 200);
+ assert.ok(comboModel);
+ assert.equal(comboModel.context_length, 64000, "manual context_length should override auto-calc");
+});