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(reasoning): preserve chat effort and protocol labels (#1550)

Randi78531fe033bf
diff --git a/CHANGELOG.md b/CHANGELOG.md
index dfe2be208a..004677c4b5 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,10 @@
## [Unreleased]
+### 🐛 Bug Fixes
+
+- **fix(providers):** Preserve OpenAI Chat Completions `reasoning_effort` through assistant-prefill requests and label OpenAI request protocols explicitly as `OpenAI-Chat` or `OpenAI-Responses`.
+
---
## [3.7.0] — 2026-04-23
diff --git a/open-sse/services/provider.ts b/open-sse/services/provider.ts
index ae454e821b..56a8d51509 100644
--- a/open-sse/services/provider.ts
+++ b/open-sse/services/provider.ts
@@ -86,17 +86,6 @@ function buildAnthropicCompatibleUrl(baseUrl) {
// contain max_tokens or Claude model names.
export function detectFormatFromEndpoint(body, endpointPath = "") {
const path = String(endpointPath || "");
- const hasInputField =
- body &&
- typeof body === "object" &&
- Object.prototype.hasOwnProperty.call(body, "input") &&
- body.input !== undefined;
- const hasResponsesSpecificFields =
- body &&
- typeof body === "object" &&
- (body.max_output_tokens !== undefined ||
- body.previous_response_id !== undefined ||
- body.reasoning !== undefined);
if (/\/responses(?=\/|$)/i.test(path) || /^responses(?=\/|$)/i.test(path)) {
return "openai-responses";
@@ -110,9 +99,6 @@ export function detectFormatFromEndpoint(body, endpointPath = "") {
/\/(?:chat\/completions|completions)(?=\/|$)/i.test(path) ||
/^(?:chat\/completions|completions)(?=\/|$)/i.test(path)
) {
- if (hasInputField || hasResponsesSpecificFields) {
- return "openai-responses";
- }
return "openai";
}
@@ -400,11 +386,10 @@ export function hasThinkingConfig(body) {
}
// Normalize thinking config based on last message role
-// - If lastMessage is not user → remove thinking config
-// - If lastMessage is user AND has thinking config → keep it (force enable)
+// - If lastMessage is not user → remove Claude/Gemini-style thinking config
+// - Keep OpenAI Chat Completions reasoning_effort as a request-level option.
export function normalizeThinkingConfig(body) {
if (!isLastMessageFromUser(body)) {
- delete body.reasoning_effort;
delete body.thinking;
}
return body;
diff --git a/src/shared/components/RequestLoggerDetail.tsx b/src/shared/components/RequestLoggerDetail.tsx
index 5b60f4dba9..f599f1d87e 100644
--- a/src/shared/components/RequestLoggerDetail.tsx
+++ b/src/shared/components/RequestLoggerDetail.tsx
@@ -2,9 +2,9 @@
import { useState, useEffect } from "react";
import {
- PROTOCOL_COLORS,
PROVIDER_COLORS,
getHttpStatusStyle as getStatusStyle,
+ getProtocolColor,
} from "@/shared/constants/colors";
import { formatDuration, formatApiKeyLabel } from "@/shared/utils/formatting";
@@ -57,12 +57,7 @@ export default function RequestLoggerDetail({ log, detail, loading, onClose, onC
const statusStyle = getStatusStyle(log.status);
const protocolKey = log.sourceFormat || log.provider;
- const protocol = PROTOCOL_COLORS[protocolKey] ||
- PROTOCOL_COLORS[log.provider] || {
- bg: "#6B7280",
- text: "#fff",
- label: (protocolKey || log.provider || "-").toUpperCase(),
- };
+ const protocol = getProtocolColor(protocolKey, log.provider);
const providerColor = PROVIDER_COLORS[log.provider] || {
bg: "#374151",
text: "#fff",
diff --git a/src/shared/components/RequestLoggerV2.tsx b/src/shared/components/RequestLoggerV2.tsx
index f5047b4640..343c12b656 100644
--- a/src/shared/components/RequestLoggerV2.tsx
+++ b/src/shared/components/RequestLoggerV2.tsx
@@ -6,9 +6,9 @@ import Card from "./Card";
import RequestLoggerDetail from "./RequestLoggerDetail";
import { copyToClipboard } from "@/shared/utils/clipboard";
import {
- PROTOCOL_COLORS,
PROVIDER_COLORS,
getHttpStatusStyle as getStatusStyle,
+ getProtocolColor,
} from "@/shared/constants/colors";
import {
formatTime,
@@ -706,12 +706,7 @@ export default function RequestLoggerV2() {
{sortedLogs.map((log) => {
const statusStyle = getStatusStyle(log.status);
const protocolKey = log.sourceFormat || log.provider;
- const protocol = PROTOCOL_COLORS[protocolKey] ||
- PROTOCOL_COLORS[log.provider] || {
- bg: "#6B7280",
- text: "#fff",
- label: (protocolKey || log.provider || "-").toUpperCase(),
- };
+ const protocol = getProtocolColor(protocolKey, log.provider);
const compatLabel = getProviderDisplayLabel(log.provider, providerNodes);
const providerColor = PROVIDER_COLORS[log.provider] || {
bg: "#374151",
diff --git a/src/shared/constants/colors.ts b/src/shared/constants/colors.ts
index 356508b25b..bb51630885 100644
--- a/src/shared/constants/colors.ts
+++ b/src/shared/constants/colors.ts
@@ -38,6 +38,15 @@ export const PROTOCOL_COLORS = {
bypass: { bg: "#6B7280", text: "#fff", label: "Bypass" },
};
+const PROTOCOL_KEY_ALIASES = {
+ "openai-chat": "openai",
+ "openai-response": "openai-responses",
+};
+
+function normalizeProtocolKey(protocol) {
+ return PROTOCOL_KEY_ALIASES[protocol] || protocol;
+}
+
// ═══════════════════════════════════════════
// Proxy Type Colors (ProxyLogger)
// ═══════════════════════════════════════════
@@ -134,3 +143,21 @@ export function getProviderColor(provider) {
}
);
}
+
+/**
+ * Get default fallback for a protocol color lookup.
+ * @param {string} protocol - Protocol key
+ * @param {string} fallbackProvider - Provider key to use as a secondary protocol key
+ * @returns {{ bg: string, text: string, label: string }}
+ */
+export function getProtocolColor(protocol, fallbackProvider) {
+ const normalized = normalizeProtocolKey(protocol);
+ return (
+ PROTOCOL_COLORS[normalized] ||
+ PROTOCOL_COLORS[fallbackProvider] || {
+ bg: "#6B7280",
+ text: "#fff",
+ label: (protocol || fallbackProvider || "-").toUpperCase(),
+ }
+ );
+}
diff --git a/tests/unit/chatcore-translation-paths.test.ts b/tests/unit/chatcore-translation-paths.test.ts
index 0977d807e6..dc7011103a 100644
--- a/tests/unit/chatcore-translation-paths.test.ts
+++ b/tests/unit/chatcore-translation-paths.test.ts
@@ -797,6 +797,55 @@ test("chatCore strips unsupported reasoning params and caps provider token field
assert.equal(call.body.max_completion_tokens, 16384);
});
+test("chatCore preserves reasoning_effort for assistant-prefill OpenAI-compatible requests", async () => {
+ const { call, result } = await invokeChatCore({
+ provider: "openai-compatible-aio",
+ model: "glm-5.1",
+ endpoint: "/v1/chat/completions",
+ body: {
+ model: "aio/glm-5.1",
+ messages: [
+ { role: "user", content: "draft the answer" },
+ { role: "assistant", content: "<thinking>" },
+ ],
+ reasoning_effort: "xhigh",
+ stream: true,
+ },
+ responseFormat: "openai",
+ });
+
+ assert.equal(result.success, true);
+ assert.equal(call.body.model, "glm-5.1");
+ assert.equal(call.body.reasoning_effort, "xhigh");
+});
+
+test("chatCore logs chat completions endpoint as OpenAI protocol", async () => {
+ const { call, result } = await invokeChatCore({
+ provider: "openrouter",
+ model: "deepseek/deepseek-v4-pro",
+ endpoint: "/v1/chat/completions",
+ body: {
+ model: "openrouter/deepseek/deepseek-v4-pro",
+ messages: [{ role: "user", content: "Human: Hi" }],
+ temperature: 1,
+ max_tokens: 64000,
+ stream: false,
+ presence_penalty: 0,
+ frequency_penalty: 0,
+ top_p: 0.9,
+ },
+ responseFormat: "openai",
+ });
+
+ assert.equal(result.success, true);
+ assert.equal(call.body.model, "deepseek/deepseek-v4-pro");
+
+ const logEntry = await waitFor(getLatestCallLog);
+ assert.ok(logEntry, "expected call log to be persisted");
+ assert.equal(logEntry.path, "/v1/chat/completions");
+ assert.equal(logEntry.sourceFormat, FORMATS.OPENAI);
+});
+
test("chatCore surfaces translation errors with explicit status codes", async () => {
register(
FORMATS.OPENAI_RESPONSES,
diff --git a/tests/unit/plan3-p0.test.ts b/tests/unit/plan3-p0.test.ts
index 51a9d62911..5995642f0a 100644
--- a/tests/unit/plan3-p0.test.ts
+++ b/tests/unit/plan3-p0.test.ts
@@ -402,12 +402,13 @@ test("detectFormat identifies OpenAI Responses by max_output_tokens without inpu
assert.equal(format, FORMATS.OPENAI_RESPONSES);
});
-test("detectFormatFromEndpoint forces OpenAI for /v1/chat/completions", () => {
+test("detectFormatFromEndpoint uses chat completions endpoint for OpenAI chat protocol", () => {
const format = detectFormatFromEndpoint(
{
- model: "cc/claude-opus-4-6",
+ model: "test-model",
messages: [{ role: "user", content: "hi" }],
- max_tokens: 16,
+ input: "ignored for endpoint protocol detection",
+ max_output_tokens: 16,
stream: false,
},
"/v1/chat/completions"
diff --git a/tests/unit/provider-service.test.ts b/tests/unit/provider-service.test.ts
index bfb075ed26..e6a4c229b9 100644
--- a/tests/unit/provider-service.test.ts
+++ b/tests/unit/provider-service.test.ts
@@ -114,7 +114,7 @@ test("Unknown providers fall back to bearer auth and OpenAI format", () => {
assert.equal(getTargetFormat("custom-provider"), "openai");
});
-test("thinking config is removed when the last message is not from the user", () => {
+test("native thinking config is removed when the last message is not from the user", () => {
const assistantLast = {
messages: [
{ role: "user", content: "hi" },
@@ -134,7 +134,7 @@ test("thinking config is removed when the last message is not from the user", ()
assert.equal(isLastMessageFromUser({ messages: [] }), true);
assert.equal(isLastMessageFromUser(assistantLast), false);
assert.equal(hasThinkingConfig(userLast), true);
- assert.equal("reasoning_effort" in normalized, false);
+ assert.equal(normalized.reasoning_effort, "high");
assert.equal("thinking" in normalized, false);
assert.equal(normalizeThinkingConfig(userLast).reasoning_effort, "medium");
});
diff --git a/tests/unit/request-log-detail-layout.test.ts b/tests/unit/request-log-detail-layout.test.ts
index a6d2dd7eb3..00166ed8b2 100644
--- a/tests/unit/request-log-detail-layout.test.ts
+++ b/tests/unit/request-log-detail-layout.test.ts
@@ -6,6 +6,46 @@ import { renderToStaticMarkup } from "react-dom/server";
const { default: RequestLoggerDetail } =
await import("../../src/shared/components/RequestLoggerDetail.tsx");
+function renderDetailWithSourceFormat(sourceFormat: string) {
+ return renderToStaticMarkup(
+ React.createElement(RequestLoggerDetail, {
+ log: {
+ status: 200,
+ method: "POST",
+ path: "/v1/chat/completions",
+ timestamp: "2026-04-09T21:27:08.000Z",
+ duration: 2500,
+ provider: "openrouter",
+ sourceFormat,
+ model: "deepseek/deepseek-v4-pro",
+ requestedModel: "openrouter/deepseek/deepseek-v4-pro",
+ cacheSource: "upstream",
+ tokens: {
+ in: 10,
+ out: 2,
+ cacheRead: null,
+ cacheWrite: null,
+ reasoning: null,
+ },
+ },
+ detail: {
+ requestedModel: "openrouter/deepseek/deepseek-v4-pro",
+ cacheSource: "upstream",
+ tokens: {
+ in: 10,
+ out: 2,
+ cacheRead: null,
+ cacheWrite: null,
+ reasoning: null,
+ },
+ },
+ loading: false,
+ onClose: () => {},
+ onCopy: async () => true,
+ })
+ );
+}
+
test("request log detail splits token badges into input and output groups", () => {
const html = renderToStaticMarkup(
React.createElement(RequestLoggerDetail, {
@@ -70,6 +110,7 @@ test("request log detail splits token badges into input and output groups", () =
assert.equal(modelLabelIndex < requestedModelLabelIndex, true);
assert.notEqual(html.indexOf(">Cache Source<"), -1);
assert.notEqual(html.indexOf(">Semantic (OmniRoute)<"), -1);
+ assert.notEqual(html.indexOf(">OpenAI-Chat<"), -1);
assert.match(
html,
@@ -77,3 +118,11 @@ test("request log detail splits token badges into input and output groups", () =
);
assert.match(html, /data-testid="token-group-output"[\s\S]*Total Out: 42[\s\S]*Reasoning: N\/A/);
});
+
+test("request log detail labels OpenAI protocol variants explicitly", () => {
+ const chatHtml = renderDetailWithSourceFormat("openai");
+ const responsesHtml = renderDetailWithSourceFormat("openai-responses");
+
+ assert.notEqual(chatHtml.indexOf(">OpenAI-Chat<"), -1);
+ assert.notEqual(responsesHtml.indexOf(">OpenAI-Responses<"), -1);
+});