Global API Benchmark

Latency and throughput benchmarks for AI API models accessible through the Global API gateway. Measure TTFT, tokens/second, and cost efficiency from 6 global regions.

Quick Benchmark

from openai import OpenAI
import time

client = OpenAI(
    base_url="https://global-apis.com/v1",
    api_key="your-global-api-key",
)

models = ["deepseek-ai/DeepSeek-V4-Flash", "qwen/qwen3-32b", "moonshot/kimi-k2.5"]
prompt = "Write a Python function to sort a list of dictionaries by multiple keys."

for model in models:
    start = time.time()
    response = client.chat.completions.create(
        model=model, messages=[{"role": "user", "content": prompt}], max_tokens=200
    )
    elapsed = (time.time() - start) * 1000
    tokens = response.usage.completion_tokens
    print(f"{model}: {elapsed:.0f}ms, {tokens/elapsed*1000:.1f} tok/s")

Latest Results (May 2026)

Speed Benchmark (TTFT in ms, lower is better)

Model	US East	US West	EU	Asia	Australia
DeepSeek V4 Flash	420	450	480	320	510
Qwen3-32B	510	530	560	380	590
GPT-4o	680	710	740	850	760
Kimi K2.5	560	590	620	410	650
GLM-5	530	560	590	390	620

Throughput (tokens/second, higher is better)

Model	US East	US West	EU	Asia	Australia
DeepSeek V4 Flash	85.2	82.1	79.5	92.3	78.4
Qwen3-32B	72.1	69.8	67.2	78.6	65.9
GPT-4o	45.3	43.1	41.8	48.2	40.5
Kimi K2.5	60.3	58.1	55.7	65.4	53.2

Streaming vs Non-Streaming

Model	Non-Stream TTFT	Stream TTFT	Improvement
DeepSeek V4 Flash	420ms	180ms	2.3x faster
Qwen3-32B	510ms	240ms	2.1x faster
GPT-4o	680ms	350ms	1.9x faster

Why Global API?

• One endpoint, 184+ models — test without signing up for each provider
• Same latency as direct — no proxy slowdown, direct access to official APIs
• Pay with PayPal — no Chinese bank account needed
• Never-expiring credits — no monthly subscription traps

Run All Benchmarks

pip install openai tabulate
python run_all_benchmarks.py --api-key your-global-api-key

Links

• Global API — one API key for all models
• Global API Pricing — compare model costs
• Global API Status — real-time uptime monitoring