Model v4 Now Available

Foundation models for
structured reasoning

Extract entities, classify intent, and generate schema-constrained outputs from unstructured text. 200K context window. Sub-200ms P95 latency.

Try the API Read Docs

app.ts config.ts

Response — 143ms

See it in action

Send a request, get structured data back. No post-processing needed.

request.ts

                response.json
              
                89ms
              
              tokens: 312 in / 47 out
              |
              stop: end_turn

Core capabilities

What the model does well, and what it was designed for.

200K Context Window

Process entire codebases, legal documents, or research papers in a single request. No chunking, no lost context.

Input and output windows are independently configurable per request.

Input tokens187,429

Output tokens2,048

Latency (P95)187ms

Structured Outputs

Define a JSON Schema and get guaranteed-valid responses. No post-processing, no regex parsing.

Works with nested objects, arrays, enums, and recursive schemas.

output.json

1{
2  "sentiment": "positive",
3  "confidence": 0.94,
4  "entities": [
5    {
6      "text": "Nexus AI",
7      "type": "ORG"
8    }
9  ]
10}

Enterprise Security

SOC 2 Type II certified. Data encrypted at rest and in transit. Zero data retention on API requests by default.

Custom data residency available for regulated industries.

SOC 2Type II

HIPAACompliant

GDPRReady

99.9%Uptime SLA

Function calling

Define tools as JSON Schema, and the model decides when and how to invoke them. Supports parallel and sequential tool use.

tools.ts types.ts

1const tools = [{

2 name: "get_weather",

3 description: "Get current weather for a location",

4 parameters: {

5 type: "object",

6 properties: {

7 location: { type: "string" },

8 unit: { type: "string", enum: ["celsius", "fahrenheit"] },

9 },

10 },

11}];

13const response = await nexus.complete({

14 model: "nexus-v4",

15 messages: [{ role: "user", content: "Weather in Tokyo?" }],

16 tools,

17});

19// response.tool_calls[0].function.name → "get_weather"

20// response.tool_calls[0].function.arguments → { location: "Tokyo" }

Safety & Alignment

Every model undergoes adversarial evaluation, bias benchmarking, and alignment tuning before release. Published model cards include failure modes and known limitations.

Red Teaming

External adversarial testing by independent security researchers before each model release.

Bias Evaluation

Continuous monitoring across demographic axes with published evaluation results and methodology.

Content Filtering

Multi-layer output filtering with configurable sensitivity for enterprise needs.

Usage-based pricing

Pay only for what you use. No seat licenses, no minimums.

Model	Input / 1M tokens	Output / 1M tokens	Context
nexus-v4	$[N]	$[N]	200K
nexus-v4-mini	$[N]	$[N]	128K
nexus-embed	$[N]	—	8K
nexus-vision	$[N]	$[N]	128K + images

Volume discounts available above $[N]/month. Contact sales

Foundation models forstructured reasoning