LLM API benchmark
Run the same five prompts against any compatible endpoint, then compare the delay before output starts, generation speed, and total response time.
What Speedtest measures
LMSpeed sends five standardized prompts to the model and records each response from the first token to completion. This separates connection and queue delay from the model's sustained generation speed.
Why this is useful
Use comparable results to spot slow relay routes, evaluate providers before integration, and choose an endpoint that feels responsive to real users.
A live sample of recent community benchmarks for comparing provider latency and output speed.
| Time | Provider | Model | Speed | Latency |
|---|---|---|---|---|
| Aug 15, 03:05 AM | mishradev.com | claude-sonnet-5 | 85.74t/s | 2.74s |
| Aug 14, 06:31 PM | open.bigmodel.cn | glm-5.3 | 420.40t/s | 27.86s |
| Aug 14, 02:55 PM | integrate.api.nvidia.com | stepfun-ai/step-3.7-flash | 267.40t/s | 37.55s |
| Aug 14, 02:55 PM | integrate.api.nvidia.com | meta/muse-glimmer-30b | 118.96t/s | 4.57s |
| Aug 14, 02:53 PM | integrate.api.nvidia.com | meta/muse-glimmer-30b | 129.74t/s | 5.10s |
Start with model score lookup, then compare per-token pricing, five-round speed tests, health checks, and safety probes before an API reaches production.
Search a model name from the homepage and jump into benchmark pages with AA score, coding, math, and model metadata.
Compare per-token pricing across 100+ providers, find cheaper APIs for each model, and track free tiers and credits.
Run a five-round benchmark with standardized prompts to measure first-token latency, output throughput, and response time.
Audit any OpenAI-compatible API for model authenticity, hidden prompts, instruction tampering, stream integrity, and error leakage, then share a plain-language report.
Enter a base URL, API key, and model ID to test official providers, proxies, relays, or self-hosted endpoints.
Review first-token latency, output throughput, total duration, health, and recent probe signals to judge stability.
How LMSpeed handles model benchmarks, provider pricing, speed tests, and API audits.
Use the model search on the homepage or open the Model Performance leaderboard. Search by model name to find the model detail page, where LMSpeed connects benchmark signals such as AA score, coding, and math with model metadata.
LMSpeed aggregates per-token pricing from 100+ API providers. Visit any model page to see a side-by-side pricing comparison table showing input and output rates per million tokens, so you can find the cheapest provider for each model.
Many providers offer free API tiers or credits for popular models like DeepSeek, Gemini, and Llama. Check our Free LLM API directory for a complete list of models with free access, including speed benchmarks for each free provider.
LMSpeed employs a five-round continuous stress testing mechanism with standardized prompts. Token calculations are performed accurately using tiktoken, measuring output throughput (tokens per second) and first-token latency.
It sends multiple safety probes to an OpenAI-compatible endpoint to check whether the model identity matches, hidden system prompts are injected, user instructions are rewritten, streaming responses stay intact, and errors leak sensitive implementation details. The result includes a risk score and a shareable report. API keys are only used for that audit and are not written to public reports.
Use our performance leaderboards and model detail pages to visually compare API speed benchmarks across providers. The system ranks providers by throughput, latency, and health, helping you choose the fastest and most reliable API.
Provider pages and the health leaderboard already show recent health checks, probe latency, success or failure status, and stability rankings. Broader continuous monitoring and alerting will keep expanding.
