Frontier models, fair price, no data sharing.
Access the latest open models at a reasonable cost, without handing over your data, through the NaN community.
nan.builders →// opendata
The real numbers
behind the platform.
Real usage data from our inference platform, aggregated and anonymized. No prompts, no content; just request-level counters.
Updated weekly · 30 Aug 20261.1T
Tokens processed
cumulative
96.9M
Requests served
cumulative
20
Active models
in production
// models
Tokens by model.
Cumulative tokens processed per model. One open model carries most of the load, and the full stack is always available.
01 Qwen 3.6 48.8%
02 DeepSeek V4-Flash 21.6%
03 MiMo V2.5 14.2%
04 deepseek-v4-flash-0731 5.5%
05 glm5.2 5%
06 Gemma 4 1.9%
07 qwen3.8-flash 1.4%
08 glm5.3-flash 1%
09 Qwen3 Embedding 0.6%
10 Qwen3 Coder <0.1%
11 claude-opus-5 <0.1%
12 claude-fable-5 <0.1%
13 deepseek <0.1%
14 claude-haiku-4-5 <0.1%
15 Whisper Large v3 <0.1%
16 gpt-5.6-luna <0.1%
17 claude-sonnet-5 <0.1%
18 gemini-3.5-flash-lite <0.1%
// tokens
Input vs. output.
Inference here is overwhelmingly read-heavy, long prompts, retrieval and context, with a thin slice of generated tokens.
Input · prompt 1.1 T 97.8%
Output · generated 23.3 B 2.2%
// usage
Tokens per day.
Daily tokens processed over the last 90 days, peaking at 20.2 B/day.
90 days agotoday
// beyond text
Speech & reranking.
The stack is more than LLMs, transcription, synthesis and reranking run on the same API.
Text-to-speech Kokoro 92.3 K requests
Reranking Qwen3 Reranker 50.5 K requests
// clients · last 7 days
How teams connect.
Drop-in OpenAI compatibility in the wild, the official SDK and OpenCode account for the vast majority of traffic.
// geography · last 7 days
Where requests come from.
70.1% of traffic originates inside the EU, the audience this infrastructure is built for.
Spain 37.6%
AT 12.9%
Colombia 10.7%
Germany 9.9%
Mexico 7.9%
United States 6.2%
France 4.5%
Finland 3.9%
Argentina 2.9%
Ireland 0.9%
Netherlands 0.6%
Chile 0.4%
Others 1.7%
// performance
Latency & throughput.
Median time to first token and sustained throughput per model, measured on 30 Aug 2026.
Model TTFT p50 Throughput
Qwen 3.6 n/a 30648 rpm
DeepSeek V4-Flash n/a 29989 rpm
MiMo V2.5 n/a 27761 rpm
Gemma 4 n/a 27743 rpm
glm5.3-flash n/a 3712 rpm
qwen3.8-flash n/a 1335 rpm
Kokoro n/a 47.7 rpm
Whisper Large v3 n/a 30.8 rpm
Qwen3 Embedding n/a 14.2 rpm
Qwen3 Reranker n/a 12.9 rpm
glm5.2 n/a 9.2 rpm
claude-opus-5 n/a n/a
TTFT p50 = median time to first token · Throughput = sustained requests per minute.
// who it's for
Two ways to run on this stack.
These numbers come from real workloads across the community and private deployments alike.
Private, dedicated inference with SLAs.
Dedicated infrastructure, support and contractual SLAs, flat rate, EU data, OpenAI-compatible.
see_pricing →Methodology. Figures are aggregated, anonymized counters collected at the request level. Helmcode keeps zero logs, no prompt or completion content is ever stored. Cumulative metrics span the platform's lifetime; windowed metrics are labelled per section.
// get started
START BURNING TOKENS
Skip the AI infra work. Deploy your first private inference endpoint today.
Flat rate. EU data. OpenAI API compatible.