🏪 LM Studio Model Benchmark

Cold-start + warm inference comparison · Local GPU Server · Feb 6, 2026

Cold Start (model load + inference)
Warm Response (routing test)
Prompt Injection Test

⏱ Cold Start Time (lower is better)

0s10s20s30s40s50s
Qwen3-VL-8B 👑 WINNER
12.4s
12.4s
x-coder-rl-qwen3-8b
19.2s
19.2s
Qwen2-VL-7B
29.9s
29.9s
Qwen2.5-VL-7B
41.7s
41.7s
uigen-t2-7b
44.0s
44.0s
InternVL3.5-8B 🔥 FASTEST
7.7s
7.7s

⚡ Warm Response Time (lower is better)

0s5s10s15s20s25s
Qwen3-VL-8B 👑
1.3s
1.3s
Qwen2-VL-7B
0.96s
0.96s
Qwen2.5-VL-7B
4.3s
4.3s
x-coder-rl-qwen3-8b
5.4s (no output)
N/A
uigen-t2-7b
25.0s
25.0s
InternVL3.5-8B 🔥
0.45s
0.45s

🎯 Quality Scores

🏆 Qwen3-VL-8B

Greeting★★★★☆
Routing★★★★★
Injection Resist★★★★☆
Cold Start12.4s
Warm Avg~1s
Overall4.3 / 5

Qwen2-VL-7B-Instruct

Greeting★★★☆☆
Routing★★☆☆☆
Injection Resist★★★☆☆
Cold Start29.9s
Warm Avg~1s
Overall2.7 / 5

Qwen2.5-VL-7B

Greeting★★★☆☆
Routing★★★☆☆
Injection Resist★★☆☆☆
Cold Start41.7s
Warm Avg4.3s
Overall2.7 / 5

uigen-t2-7b

Greeting★★☆☆☆
Routing★★★★☆
Injection Resist★★★★☆
Cold Start44.0s
Warm Avg25.0s
Overall3.3 / 5

❌ x-coder-rl-qwen3-8b

Greeting☆☆☆☆☆
Routing☆☆☆☆☆
Injection Resist★☆☆☆☆
IssueNo user output (think-only)
Overall0 / 5

🔥 InternVL3.5-8B

Greeting★★★★☆
Routing★★★★☆
Injection Resist★★★★★
Cold Start7.7s
Warm Avg~0.4s
Overall4.3 / 5

🏆 Verdict

Two-way tie: Qwen3-VL-8B and InternVL3.5-8B both score 4.3/5 quality. InternVL3.5 wins on speed (7.7s cold, 0.4s warm) while Qwen3-VL-8B has slightly better routing. InternVL3.5 has perfect injection resistance (5/5). Either is production-ready with a hardened system prompt.