diff --git a/Cargo.lock b/Cargo.lock
index 6cffadc..44855c5 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -869,12 +869,6 @@ dependencies = [
  "syn 1.0.109",
 ]
 
-[[package]]
-name = "diff"
-version = "0.1.13"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "56254986775e3233ffa9c4d7d3faaf6d36a2c09d30b20687e9f88bc8bafc16c8"
-
 [[package]]
 name = "digest"
 version = "0.10.7"
@@ -1219,21 +1213,6 @@ version = "0.3.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d2fabcfbdc87f4758337ca535fb41a6d701b65693ce38287d856d1674551ec9b"
 
-[[package]]
-name = "goldie"
-version = "0.4.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fd1d4b95ae93c6d91591a2998aa7363113e51130ede293b3c92ac89b63e13914"
-dependencies = [
- "anyhow",
- "once_cell",
- "pretty_assertions",
- "serde",
- "serde_json",
- "upon",
- "yansi",
-]
-
 [[package]]
 name = "h2"
 version = "0.3.19"
@@ -2335,16 +2314,6 @@ version = "0.2.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5b40af805b3121feab8a3c29f04d8ad262fa8e0561883e7653e024ae4479e6de"
 
-[[package]]
-name = "pretty_assertions"
-version = "1.4.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "af7cee1a6c8a5b9208b3cb1061f10c0cb689087b3d8ce85fb9d2dd7a29b6ba66"
-dependencies = [
- "diff",
- "yansi",
-]
-
 [[package]]
 name = "prettyplease"
 version = "0.1.25"
@@ -3146,7 +3115,6 @@ dependencies = [
  "clap",
  "ctranslate2-bindings",
  "futures",
- "goldie",
  "http-api-bindings",
  "hyper",
  "lazy_static",
@@ -4093,17 +4061,6 @@ version = "0.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "39ec24b3121d976906ece63c9daad25b85969647682eee313cb5779fdd69e14e"
 
-[[package]]
-name = "upon"
-version = "0.5.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b77ce40602cb1a7dfcdd6336f6d8baa2803c898aafbc0d46156b59727f2e7135"
-dependencies = [
- "serde",
- "unicode-ident",
- "unicode-width",
-]
-
 [[package]]
 name = "url"
 version = "2.3.1"
@@ -4592,12 +4549,6 @@ dependencies = [
  "libc",
 ]
 
-[[package]]
-name = "yansi"
-version = "0.5.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "09041cd90cf85f7f8b2df60c646f853b7f535ce68f85244eb6731cf89fa498ec"
-
 [[package]]
 name = "zip"
 version = "0.6.6"
diff --git a/Makefile b/Makefile
index 25e7ca1..9aa672c 100644
--- a/Makefile
+++ b/Makefile
@@ -1,8 +1,9 @@
-smoke:
-	k6 run tests/*.smoke.js
-
 loadtest:
+ifdef TABBY_API_HOST
 	k6 run tests/*.loadtest.js
+else 
+	$(error TABBY_API_HOST is undefined)
+endif
 
 fix:
 	cargo clippy --fix --allow-dirty --allow-staged && cargo +nightly fmt
diff --git a/tests/default.loadtest.js b/tests/default.loadtest.js
index 3280376..f961632 100644
--- a/tests/default.loadtest.js
+++ b/tests/default.loadtest.js
@@ -1,23 +1,28 @@
 import http from "k6/http";
-import { check, group, sleep } from "k6";
+import { check, group, sleep, abortTest } from "k6";
 
 export const options = {
   stages: [
-    { duration: "5s", target: 10 }, // simulate ramp-up of traffic from 1 to 10 users over 30s.
-    { duration: "30s", target: 10 }, // stay at 10 users for 10 minutes
-    { duration: "5s", target: 0 }, // ramp-down to 0 users
+    { duration: "5s", target: 8 },
+    { duration: "20s", target: 8 },
+    { duration: "5s", target: 0 },
   ],
   hosts: {
-    "api.tabbyml.com": __ENV.TABBY_API_HOST || "localhost:5000",
+    "api.tabbyml.com": __ENV.TABBY_API_HOST,
   },
+  // Below thresholds are tested against TabbyML/StarCoder-1B served by NVIDIA T4 GPU.
   thresholds: {
-    http_req_duration: ["p(99)<1000"], // 99% of requests must complete below 1000ms
+    http_req_failed: ['rate<0.01'], // http errors should be less than 1%
+    http_req_duration: ["med<400", "avg<1500", "p(95)<3000"],
   },
 };
 
 export default () => {
   const payload = JSON.stringify({
-    prompt: "def binarySearch(arr, left, right, x):\n    mid = (left +",
+    language: "python",
+    segments: {
+      prefix: "def binarySearch(arr, left, right, x):\n    mid = (left +"
+    },
   });
   const headers = { "Content-Type": "application/json" };
   const res = http.post("http://api.tabbyml.com/v1/completions", payload, {
diff --git a/tests/default.smoke.js b/tests/default.smoke.js
deleted file mode 100644
index 30697d9..0000000
--- a/tests/default.smoke.js
+++ /dev/null
@@ -1,24 +0,0 @@
-import http from "k6/http";
-import { check, sleep } from "k6";
-
-export const options = {
-    stages: [
-        {duration: '3s', target: 5},
-    ],
-    hosts: {
-        'api.tabbyml.com': __ENV.TABBY_API_HOST || "localhost:5000"
-    },
-};
-const SLEEP_DURATION = 1;
-
-export default function () {
-  const payload = JSON.stringify({
-    prompt: "def binarySearch(arr, left, right, x):\n    mid = (left +",
-  });
-  const headers = { "Content-Type": "application/json" };
-  const res = http.post("http://api.tabbyml.com/v1/completions", payload, {
-    headers,
-  });
-  check(res, { success: (r) => r.status === 200 });
-  sleep(SLEEP_DURATION)
-}