Midas Engine

The core AI engine powering Midas Code.

Midas Engine is the model layer beneath the Midas Code API. Built from the ground up for code generation with a 200K token context window and sub-second latency.

Read the Docs Get API Access

Model Performance

94.7%

HumanEval

Pass@1

91.2%

MBPP

Pass@1

200K

Context

Token window

<800ms

Latency

Median p50

Built for code, not repurposed for it

Midas-2 Model

Our latest code-focused model, trained on 500B tokens of high-quality code across 40+ languages and frameworks.

Context Architecture

200K token context window. Pass entire codebases, documentation, and examples. The engine understands your full project.

Low Latency

Median latency under 800ms for standard generations. Streaming available for interactive use cases.

Technical capabilities

200K token context window for large codebase understanding
Fine-tuned on code correctness, not just code completion
Multi-turn context for iterative refinement
Structured output mode for JSON, YAML, and schema generation
Tool use for reading files, running tests, and fetching docs
Function calling for integration with existing systems

Structured output mode

{
  "model": "midas-2",
  "output_format": "json",
  "schema": {
    "type": "object",
    "properties": {
      "code": { "type": "string" },
      "language": { "type": "string" },
      "tests": {
        "type": "array",
        "items": { "type": "string" }
      }
    }
  }
}

Access Midas Engine via the API

Get API Access Read the Docs

The same engine that powers Midas Code is available directly via our REST API. Build anything.