dsh-livebench-panel 0.1.9 → 0.1.10

package/lib/index.js

@@ -381,6 +381,14 @@ function writeGeneratedModelConfig(layout, YAML, { displayName, modelId, reasoni
 			// note: reasoning_effort is an OpenAI-style knob and is intentionally
 			// not forwarded on the anthropic protocol path.
 		}
+	} else if (protocol === "openai_responses") {
+		// OpenAI Responses API proxy: provider name matches
+		// get_api_function('openai_responses') → chat_completion_openai_responses,
+		// which converts reasoning_effort to reasoning:{effort} itself.
+		lines.push(`  openai_responses: ${modelId}`);
+		if (reasoningEffort) {
+			lines.push("api_kwargs:", "  default:", `    reasoning_effort: ${reasoningEffort}`);
+		}
 	} else {
 		lines.push(`  local: ${modelId}`);
 		if (reasoningEffort) {
@@ -557,12 +565,19 @@ function apply(ctx) {
 		// selects LiveBench's native anthropic client and the spawn env points
 		// the SDK at the proxy (ANTHROPIC_BASE_URL / ANTHROPIC_API_KEY).
 		const isAnthropicRoute = provider && provider.api === "anthropic-messages" && provider.baseURL;
-		if (isAnthropicRoute || hasEffortSuffix) {
+		// OpenAI Responses-API proxies (api: openai-responses) select LiveBench's
+		// openai_responses client; --api-base + LIVEBENCH_API_KEY route them the
+		// usual way, and reasoning_effort converts to reasoning:{effort} inside.
+		const isOpenAIResponsesRoute = provider
+			&& (provider.api === "openai-responses" || provider.api === "openai_responses")
+			&& provider.baseURL;
+		const needConfig = isAnthropicRoute || isOpenAIResponsesRoute || hasEffortSuffix;
+		if (needConfig) {
 			writeError = writeGeneratedModelConfig(layout, loadYaml(profileDir), {
 				displayName,
 				modelId,
 				reasoningEffort: isAnthropicRoute ? null : reasoningEffort,
-				protocol: isAnthropicRoute ? "anthropic" : "openai",
+				protocol: isAnthropicRoute ? "anthropic" : isOpenAIResponsesRoute ? "openai_responses" : "openai",
 			});
 			if (writeError) return { status: 500, payload: { ok: false, error: writeError } };
 			cliModel = displayName;
@@ -693,7 +708,8 @@ function apply(ctx) {
 				providers: providers.map(({ id, name: pname, models, api, baseURL }) => ({
 					id,
 					name: pname,
-					routable: (api === "openai-completions" || api === "anthropic-messages") && typeof baseURL === "string" && baseURL.length > 0,
+					routable: ["openai-completions", "openai-responses", "openai_responses", "anthropic-messages"].includes(api)
+						&& typeof baseURL === "string" && baseURL.length > 0,
 					baseURL: baseURL ?? null,
 					models,
 				})),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-livebench-panel",
-  "version": "0.1.9",
+  "version": "0.1.10",
   "description": "DSH web plugin: a LiveBench tab in the Trajectory view (right of 对话/轨迹). Run LiveBench evaluations against every model configured in the DeepSeek Harness — pick provider/model, category, task, release and question range from dropdowns, watch progress, and read scores in place.",
   "license": "MIT",
   "type": "module",