Open weights caught up. Compare EU-hosted inference providers.
DeepSeek, Qwen, and others now rival proprietary giants. ttft.eu measures the European inference providers that run them: time to first token, throughput, request success rates, and more.
This site isn't ready for prime time yet. We're still collecting data.
Notify me01Pricing
02Speed
Latency avg
TTFT avg
Tok/s avg
03What gets measured
- TTFTtime to first token
- Output tokens/secfixed , output capped at 512 tokens
- LatencyDNS + TCP + TLS connect time
- Success ratesuccessful requests over the selected period
- € / 1M input tokenslist price, billed in EUR
- € / 1M output tokenslist price, billed in EUR
04Why benchmarks matter
Open-weight models have democratized frontier AI, but inference infrastructure remains the wild west. Static pricing sheets ignore cold starts, throttling, and degraded time to first token during peak traffic. ttft.eu runs continuous end-to-end tests against real provider endpoints to expose true production performance.