Nebius Token Factory
Try Nebius Token Factory on us. Starter credit covers your first requests, then one key reaches all 26 Nebius Token Factory models, with the privacy posture shown on every route.
Nebius Token Factorynebius
These privacy labels describe Nebius Token Factory, the upstream model provider. Confidential requires all three: verified provider compute, end-to-end encryption and explicit zero data retention. Account and billing metadata are outside the content-retention claim.
| Provider | Nebius Token Factory |
|---|---|
| API operator country | NL Legal home of the API operator, not an inference-location guarantee. |
| Headquarters | Amsterdam, Netherlands Company newsroom identifies its headquarters. Source; reviewed 2026-10-28. |
| Routing status | Active |
| Provider website | https://nebius.com/ai-studio |
| Models | 26 public models |
| Credits routes | 26 |
| Zero data retention | yes All routes |
| Verified confidential inference | Not verified |
| Policy note | Marked ZDR via TrustedRouter's arrangement — Nebius RETAINS inputs/outputs by default (for speculative decoding); zero retention is an opt-in control, which the deployed Nebius account has enabled. Nebius does not train on customer data. Policy source |
Inference locations
Provider API declarations are also available on each model's endpoint API. Direct chat and Responses replies separate advertised availability from serving location in usage.inference_location; the actual serving region remains unknown when not reported.
usage.provider_usage.gateway_region identifies the TrustedRouter gateway, not the provider's inference location. A value such as europe-west4 does not mean the model ran in Europe. usage.inference_location.serving_region remains null when the provider has not reported it. Older gateways may return the gateway-only field as provider_usage.region.
Where the upstream GPUs execute the model, not the provider's headquarters, API ingress, or data storage. ZDR and confidential inference do not by themselves restrict geography.
- Declared countries or regions
- Global (dynamic public endpoints)
Token Factory public endpoints are Global; Finland storage of retained content does not identify the inference location. - Can locations change?
- Public-endpoint processing location can change without notice.
- Provider region pinning
- Dedicated endpoints provide a selected-region commitment with no cross-region failover; public endpoints do not.
- Pinning through TrustedRouter
- Not supported by the current TrustedRouter integration.
- Infrastructure declaration
- Global, dynamically placed public inference; no fixed-country commitment.
- Evidence
- Token Factory legal guide, section 5. Reviewed .
Provider selection alone is not a region pin. TrustedRouter's US provider filter selects company jurisdiction, not GPU location. A regional gateway hostname does not pin a downstream GPU. For a binding residency requirement, confirm the model, region, failover scope and enforced controls before sending sensitive data.
Measured performance
23 samplesContinuously sampled across Nebius Token Factory's routed models: p50 TTFT, effective throughput, and success rate. Effective throughput uses provider-reported output tokens over complete request time. Unsupported route and probe-configuration rows are separated from provider downtime. No prompt or output content stored.
| p50 TTFT | 1275 ms |
|---|---|
| Effective throughput | 78 tok/s n=3 |
| Uptime | 100.00% |
| Model | p50 TTFT | Effective throughput | Uptime | Config excluded | Availability samples |
|---|---|---|---|---|---|
| openbmb/MiniCPM-V-4_5 | 749 ms | — | 100.00% | — | 2 |
| nvidia/nemotron-3_5-lightning | 891 ms | — | 100.00% | 1 provider_error |
1 |
| zai-org/GLM-5.1 | 932 ms | — | 100.00% | — | 2 |
| z-ai/glm-5.3 | 1000 ms | 4 tok/s n=1 | 100.00% | — | 2 |
| deepseek/deepseek-v4.1-flash-1017 | 1007 ms | — | 100.00% | — | 2 |
| moonshotai/Kimi-K2.6 | 1224 ms | — | 100.00% | — | 2 |
| reflection/beam | 1275 ms | — | 100.00% | — | 1 |
| zai-org/GLM-5.2 | 1348 ms | — | 100.00% | — | 1 |
| NousResearch/Hermes-4-405B | 1375 ms | — | 100.00% | — | 3 |
| moonshotai/Kimi-K2.7-Code | 1479 ms | — | 100.00% | — | 1 |
| google/gemma-3-27b-it | 1738 ms | — | 100.00% | — | 3 |
| deepseek-ai/DeepSeek-V4-Pro | 2009 ms | — | 100.00% | — | 1 |
| nvidia/nemotron-3.5-super-vl | 2137 ms | — | 100.00% | — | 1 |
| deepseek/deepseek-v4-flash-0731 | 2793 ms | — | 100.00% | 1 provider_error |
1 |
| Qwen/Qwen3-235B-A22B-Instruct-2507 | — | — | — | 3 provider_error |
0 |
| deepseek/deepseek-v4.1-flash | — | 103 tok/s n=1 | — | — | 0 |
| google/gemma-2-2b-it | — | — | — | 1 provider_error |
0 |
| moonshotai/kimi-k3 | — | 78 tok/s n=1 | — | — | 0 |
Nebius Token Factory performance history · Full provider & model leaderboard.
Models served by Nebius Token Factory.
Each row links to pricing, provider, benchmark, and API pages for the model.
| Model | AI IQ | Context | Input | Cached input | Output |
|---|---|---|---|---|---|
MiniMaxAI/MiniMax-M3MiniMax-M3 |
IQ 114#64 | 1,048,576 | $0.3165/1M | Not published | $1.266/1M |
NousResearch/Hermes-4-405BHermes 4 405B |
— | 131,072 | $1.055/1M | Not published | $3.165/1M |
Qwen/Qwen3-235B-A22B-Instruct-2507Qwen3 235B A22B Instruct 2507 |
— | 131,072 | $0.211/1M | Not published | $0.633/1M |
Qwen/Qwen3-30B-A3B-Instruct-2507Qwen3 30B A3B Instruct 2507 |
— | 131,072 | $0.1055/1M | Not published | $0.3165/1M |
Qwen/Qwen3.5-397B-A17BQwen3.5 397B A17B |
— | 262,144 | $0.633/1M | Not published | $3.798/1M |
deepseek-ai/DeepSeek-V4-ProDeepSeek V4 Pro |
IQ 123#34 | 1,048,576 | $1.84625/1M | Not published | $3.6925/1M |
deepseek/deepseek-v4-flash-0731DeepSeek: DeepSeek V4 Flash 0731 |
— | 1,048,576 | $0.1477/1M | Not published | $0.2954/1M |
deepseek/deepseek-v4.1-flashDeepSeek: DeepSeek V4.1 Flash |
IQ 126#24 | 1,048,576 | $0.3165/1M | Not published | $1.266/1M |
deepseek/deepseek-v4.1-flash-1017DeepSeek: DeepSeek V4.1 Flash 1017 |
IQ 128#19 | 1,048,576 | $0.3165/1M | Not published | $1.266/1M |
google/gemma-3-27b-itGoogle: Gemma 3 27B |
— | 131,072 | $0.1055/1M | Not published | $0.3165/1M |
meta-models/muse-spark-openMuse Spark Open |
IQ 123#35 | 1,048,576 | $1.31875/1M | Not published | $4.48375/1M |
minimax/minimax-m3.1-flashMiniMax: MiniMax M3.1 Flash |
IQ 120#49 | 1,048,576 | $0.15825/1M | Not published | $0.633/1M |
moonshotai/Kimi-K2.6Kimi-K2.6 |
IQ 114#62 | 8,000 | $1.00225/1M | Not published | $4.22/1M |
moonshotai/Kimi-K2.7-CodeKimi-K2.7-Code |
IQ 111#77 | 262,144 | $1.00225/1M | Not published | $4.22/1M |
moonshotai/kimi-k3MoonshotAI: Kimi K3 |
IQ 127#22 | 1,048,576 | $3.165/1M | Not published | $15.825/1M |
nvidia/nemotron-3-super-120b-a12bNVIDIA: Nemotron 3 Super |
— | 262,144 | $0.3165/1M | Not published | $0.9495/1M |
nvidia/nemotron-3.5-super-vlNVIDIA: Nemotron 3.5 Super VL |
IQ 105#100 | 262,144 | $0.3165/1M | Not published | $0.9495/1M |
nvidia/nemotron-3_5-lightningNemotron-3.5-Lightning |
— | 8,000 | $0.0633/1M | Not published | $0.2532/1M |
openai/gpt-oss-120bOpenAI: gpt-oss-120b |
IQ 98#136 | 131,072 | $0.15825/1M | Not published | $0.633/1M |
openbmb/MiniCPM-V-4_5openbmb/MiniCPM-V-4_5 |
— | 8,000 | $0.69419/1M | Not published | $1.17105/1M |
qwen/qwen3.8-27bQwen: Qwen3.8 27B |
IQ 115#57 | 65,536 | $0.47475/1M | Not published | $3.165/1M |
reflection/beamReflection Beam 1.0 |
IQ 114#60 | 1,048,576 | $0.26375/1M | Not published | $1.055/1M |
z-ai/glm-5.3Z.ai: GLM 5.3 |
IQ 124#30 | 1,048,576 | $1.477/1M | Not published | $4.642/1M |
z-ai/glm-5.3-flashZ.ai: GLM 5.3 Flash |
IQ 119#50 | 1,048,576 | $0.15825/1M | Not published | $0.5275/1M |
zai-org/GLM-5.1GLM 5.1 |
IQ 112#67 | 204,800 | $1.477/1M | Not published | $4.642/1M |
zai-org/GLM-5.2GLM-5.2 |
IQ 118#53 | 1,048,576 | $1.477/1M | Not published | $4.642/1M |
Questions
Does Nebius Token Factory have zero data retention?
TrustedRouter records Nebius Token Factory as supporting provider-level zero data retention based on the policy source linked on this page. This is a provider policy claim, separate from TrustedRouter's content-stateless real-time gateway and from end-to-end confidential compute.
Is Nebius Token Factory end-to-end encrypted?
TrustedRouter does not currently mark Nebius Token Factory as end-to-end encrypted at the provider boundary. The TrustedRouter gateway is still attested, but the selected provider normally receives the request in order to run the model. Use trustedrouter/e2e for the stronger route requirement.
Which Nebius Token Factory models are available through TrustedRouter?
This page currently lists 26 public Nebius Token Factory models, with live pricing, route count, context length, measured performance when available, and links to each model's provider and benchmark pages.