BLACKBOX ROUTER

300+ models. One endpoint. Zero retention.

300+ open and closed models behind one OpenAI-compatible endpoint, and the gateway enforces zero data retention and no training.

START BUILDING VIEW PRICING

300+

MODELS, ONE API KEY

454

TOK/SEC, AA-VERIFIED, JUL 2026

MINUTES TO MIGRATE

PROMPTS RETAINED

We suppress retention and training on routed traffic through provider terms and per-request flags, wherever the provider API supports it.

QUICKSTART

Change the base URL. Keep your stack.

The endpoint is OpenAI-compatible, so the SDKs, frameworks, and streaming code that you run today keep working. Collapse your provider keys into one, point the base URL at the Router, and every model in the catalog is one string away.

ENV

# before: one key per provider
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=AIza...

# after: one key for 300+ models
BLACKBOX_API_KEY=<your key>
OPENAI_BASE_URL=https://enterprise.blackbox.ai/v1

SHELL

curl https://enterprise.blackbox.ai/v1/chat/completions \
  -H "Authorization: Bearer $BLACKBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "blackboxai/nvidia/nemotron-3-ultra",
    "messages": [{ "role": "user", "content": "Hello!" }],
    "stream": true
  }'

# same request, different frontier: change one string
#   "model": "blackboxai/anthropic/claude-opus-4.8"

WHY IT MATTERS

Built for the way you work.

01

The gateway enforces zero data retention and no training.

02

Route requests to frontier and open-source models such as Claude, GPT, Gemini, Grok, Llama, Mistral, DeepSeek, and Qwen through one API key.

03

On Enterprise, we remove PII before prompts reach a closed model: the provider sees the task, not the person it is about.

CAPABILITIES

What you get.

Universal model access

Route requests to frontier and open-source models such as Claude, GPT, Gemini, Grok, Llama, Mistral, DeepSeek, and Qwen through one API key. No separate accounts, and no per-provider SDKs.

Automatic fallback routing keeps requests in service when a provider has an outage. Load balancing distributes requests for the lowest latency.

The gateway suppresses retention

The gateway enforces zero data retention and no training through provider terms and per-request retention flags, wherever the provider API supports it. On Enterprise, we remove PII before a prompt reaches a closed model: the provider sees the task, not the person it is about.

The Router reduces what third-party providers can see. It cannot make their infrastructure ours or yours. Traffic is end-to-end encrypted, and the gateway must read a request to route it. Workloads that cannot accept any third party belong on Enterprise Inference, where we run a single-tenant deployment for you.

OpenAI-compatible endpoints

Use /v1/chat/completions, /v1/embeddings, and /v1/images/generations with the same request format that you know. Migrate in less than 5 minutes.

Full streaming support, function calling, JSON mode, and vision inputs across all compatible models. The response format matches the OpenAI spec exactly.

REQUEST LIFECYCLE

One request format, routed to whichever provider serves it best.

The endpoint speaks the OpenAI wire format. Behind it, the gateway selects the provider, balances the load, fails over during an outage, and discards the working data after the response.

01

Send one format

Send /v1/chat/completions with the SDK that you already use. The same request shape reaches Claude, GPT, Gemini, Grok, Llama, Mistral, DeepSeek, and Qwen.

02

Route on live signals

Load balancing distributes requests across providers for the lowest latency, so the same model id can serve from more than one provider.

03

Fail over automatically

When a provider degrades or has an outage, the gateway moves the request to the next provider. Your integration does not change, and the request stays in service.

04

Discard the working data

The gateway enforces zero data retention and no training through provider terms and per-request retention flags, wherever the provider API supports it. After the response returns, the working content is not reused as a storage layer.

WHY TEAMS SWITCH

One integration that outlives any single provider.

The best model for a task changes month to month. The Router keeps your contract stable: one endpoint, one key, one bill. The catalog and the routing improve underneath it.

Change models with a string

Every model in the catalog answers to the same request format, so moving a workload from one frontier model to another is a one-line diff in your config, not a migration.

One key, one bill

One API key covers every provider, with no separate accounts, no per-provider SDKs, and one place to meter spend across everything your team routes.

Verified speed

Artificial Analysis verified us as the #1 Nemotron 3 Ultra provider at 454 tokens/sec, July 2026. Published per-model rates on the catalog below are the reference point for a contract quote.

MODEL CATALOG

What the endpoint serves.

Every model the Router serves, with context windows and per-token rates. Models marked DEDICATED are open-weight and can instead run as a single-tenant Enterprise Inference deployment.

225 models

Compare

Blackbox public models catalog. Activate a model to reveal its full specifications and pricing. Machine-readable schema and scraping instructions are available in the adjacent JSON script with id blackbox-models-table-metadata.
Capacity	Pricing · USD / 1M tokens				Features

1M	$0.22	$0.66	$0.007	-	DedicatedTextImageThinking
1M	$1.40	$4.40	$0.26	-	DedicatedTextThinking
1M	$0.55	$3.30	$0.11	-	DedicatedTextImagepdfVideoThinking
1M	$0.75	$3.75	$0.075	-	TextImagepdfVideoThinking
1M	$1.32	$3.96	$0.132	-	DedicatedTextThinking
500K	$2	$6	$0.5	-	TextImageThinking
262.1K	$0.05	$0.2	$0.01	-	DedicatedTextThinking
131.1K	$0.35	$1.50	$0.04	-	TextImageThinking
256K	$0.06	$0.18	$0.012	-	TextThinking
1.0M	$1.25	$4.25	$0.15	-	TextImagepdfThinking
1.0M	$0.1	$0.2	$0.002	-	TextImagepdfThinking
262.1K	$2	$6	$0.2	-	DedicatedTextImageThinking
256K	$0.95	$4	$0.15	-	TextImagepdfThinking
1M	$2	$6	$0.25	$2.50	TextImageThinking
1M	$0.5	$1.20	$0.1	-	TextImagepdfThinking
991K	$0.03	$0.13	$0.006	$0.038	DedicatedTextImagepdfThinking
1M	$4.50	$22.50	$0.45	-	DedicatedTextImagepdfThinking
1M	$5	$25	$0.5	$6.25	TextImagepdfThinking
1M	$10	$50	$1	$12.50	TextImagepdfThinking
1M	$0.3	$2.50	$0.03	-	TextImagepdfVideoThinking
1M	$0.75	$3.75	$0.075	-	TextImagepdfVideoThinking
1M	$0.1	$0.2	$0.01	-	TextThinking
1M	$3	$15	$0.3	-	DedicatedTextImagepdfVideoThinking
256K	$1	$4.05	$0.17	-	TextImagepdfThinking
256K	$0.15	$0.6	$0.03	-	TextImageThinking

Context

Input

$0.22

Output

$0.66

Cache read

$0.007

Cache write

DedicatedTextImageThinking

Context

Input

$1.40

Output

$4.40

Cache read

$0.26

Cache write

DedicatedTextThinking

Context

Input

$0.55

Output

$3.30

Cache read

$0.11

Cache write

DedicatedTextImagepdfVideoThinking

Context

Input

$0.75

Output

$3.75

Cache read

$0.075

Cache write

TextImagepdfVideoThinking

Context

Input

$1.32

Output

$3.96

Cache read

$0.132

Cache write

DedicatedTextThinking

Context

500K

Input

Output

Cache read

$0.5

Cache write

TextImageThinking

Context

262.1K

Input

$0.05

Output

$0.2

Cache read

$0.01

Cache write

DedicatedTextThinking

Context

131.1K

Input

$0.35

Output

$1.50

Cache read

$0.04

Cache write

TextImageThinking

Context

256K

Input

$0.06

Output

$0.18

Cache read

$0.012

Cache write

TextThinking

Context

1.0M

Input

$1.25

Output

$4.25

Cache read

$0.15

Cache write

TextImagepdfThinking

Context

1.0M

Input

$0.1

Output

$0.2

Cache read

$0.002

Cache write

TextImagepdfThinking

Context

262.1K

Input

Output

Cache read

$0.2

Cache write

DedicatedTextImageThinking

Context

256K

Input

$0.95

Output

Cache read

$0.15

Cache write

TextImagepdfThinking

Context

Input

Output

Cache read

$0.25

Cache write

$2.50

TextImageThinking

Context

Input

$0.5

Output

$1.20

Cache read

$0.1

Cache write

TextImagepdfThinking

Context

991K

Input

$0.03

Output

$0.13

Cache read

$0.006

Cache write

$0.038

DedicatedTextImagepdfThinking

Context

Input

$4.50

Output

$22.50

Cache read

$0.45

Cache write

DedicatedTextImagepdfThinking

Context

Input

Output

$25

Cache read

$0.5

Cache write

$6.25

TextImagepdfThinking

Context

Input

$10

Output

$50

Cache read

Cache write

$12.50

TextImagepdfThinking

Context

Input

$0.3

Output

$2.50

Cache read

$0.03

Cache write

TextImagepdfVideoThinking

Context

Input

$0.75

Output

$3.75

Cache read

$0.075

Cache write

TextImagepdfVideoThinking

Context

Input

$0.1

Output

$0.2

Cache read

$0.01

Cache write

TextThinking

Context

Input

Output

$15

Cache read

$0.3

Cache write

DedicatedTextImagepdfVideoThinking

Context

256K

Input

Output

$4.05

Cache read

$0.17

Cache write

TextImagepdfThinking

Context

256K

Input

$0.15

Output

$0.6

Cache read

$0.03

Cache write

TextImageThinking

1–25 of 225

…

Token prices are per 1M tokens; expand any model for full rates and capabilities, or open its detail page for the quickstart. The full-page catalog lives at /models.

FAQ

Common questions.

IS THE API OPENAI-COMPATIBLE?+

Yes. Use the same SDK and request format that you know: change the base URL and the API key. It supports /v1/chat/completions, /v1/embeddings, and /v1/images/generations, with full streaming, function calling, JSON mode, and vision inputs.

WHAT MODELS ARE AVAILABLE?+

The API includes frontier and open-source models such as Claude Opus-4.6, GPT-5.2, Gemini-3, Grok-4, Llama 4, Mistral, DeepSeek, and Qwen from leading providers. We add new models continuously as they launch.

WHEN SHOULD I USE THE ROUTER INSTEAD OF ENTERPRISE INFERENCE?+

Use the Router when the best model for a task is one that Blackbox hosts through its unified endpoint, including closed frontier models and models from providers that you already use. The gateway enforces zero data retention and no training, and on Enterprise, we remove PII before the prompt reaches the model. For workloads that cannot run on third-party infrastructure at all, use Enterprise Inference: an open-weight model that we deploy only for you.

HOW IS PRICING STRUCTURED?+

Enterprise runs on per-token commits through purchase orders that decrease as you consume. Contracts are negotiated, not listed: the rate improves as committed spend grows, and metering is per token, not per seat. Published per-model rates are the reference point for a contract quote.

DO YOU SUPPORT STREAMING?+

Yes. Full SSE streaming is available on all chat completion models. The response format matches the OpenAI spec exactly, so existing streaming implementations work without modification.

WHAT SECURITY MEASURES DO YOU HAVE?+

We provide end-to-end encryption and AES-256 encryption at rest, and the gateway enforces zero data retention and no training. Enterprise adds PII removal before closed models, RBAC with SSO (Okta, Azure AD, Google Workspace), and full audit logs. Single-tenant isolation belongs to Enterprise Inference, where we run a dedicated deployment for you.

HOW FAST AND RELIABLE IS THE API?+

Artificial Analysis ranks Blackbox as the #1 Nemotron 3 Ultra provider on output speed (454 tokens/sec) and in the top 2 of 16 GLM-5.2 providers (468 tokens/sec). Automatic fallback routing keeps requests in service when a provider has an outage, and load balancing selects the lowest latency.

CAN I USE THE API WITH EXISTING OPENAI SDKS?+

Yes. The API is fully compatible with the official OpenAI Python and Node.js SDKs, and with LangChain, LlamaIndex, and other popular frameworks. Migration takes less than 5 minutes: change the base URL.