model-benchmark-gen

Generate AI model benchmark charts as SVG. Zero dependencies.

Two chart types:

• Grouped bar chart — benchmark scores per model (OpenAI/Anthropic style, dark theme)
• Log-scale scatter chart — model intelligence vs parameter count

Requires Node.js 18+.

Install

npm install model-benchmark-gen

Or use it locally from this repo — require('./src/index') works as-is, no build step.

Usage

Bar chart

const { generateChart } = require('model-benchmark-gen');

const svg = generateChart({
  title: 'Model Evaluation Suite',
  benchmarks: [
    { name: 'MMLU',     models: [{ name: 'VoidGPT 1.1', score: 67.6 }] },
    { name: 'BoolQ',    models: [{ name: 'VoidGPT 1.1', score: 85.3 }] },
    { name: 'Hellaswag', models: [
      { name: 'VoidGPT 1.1', score: 87.9 },
      { name: 'AlexLM 2.0',  score: 84.2 }
    ]}
  ]
}, 'output/benchmark.svg');
// returns the SVG string; passing a path also writes the file

Shorthand for a single benchmark:

generateChart({
  benchmarkName: 'My SWE Benchmark',
  models: [
    { name: 'Model A', percentage: 68.1 },
    { name: 'Model B', percentage: 64.3 }
  ]
}, 'output/bench.svg');

Scatter chart

const { generateScatterChart } = require('model-benchmark-gen');

generateScatterChart([
  { name: 'VoidGPT-3', params: 7,  intelligence: 82 },
  { name: 'VoidGPT-4', params: 70, intelligence: 96 }
], 'output/scatter.svg', {
  title: 'AI Model Intelligence vs Parameters',
  xLabel: 'Parameters (Billions)',
  yLabel: 'Intelligence'
});

options may also be passed as the second argument (generateScatterChart(models, { xLabel: '...' })), with outputPath inside it.

API

Function	Input	Output
generateChart(data, [outputPath])	{ title, benchmarks: [{ name, models: [{ name, score }] }] } or { benchmarkName, models: [{ name, percentage }] }	SVG string
generateScatterChart(models, [outputPath], [options])	[{ name, params, intelligence }] — params must be > 0 (log scale)	SVG string
generateColors(count)	number	array of #rrggbb colors

Scores in the bar chart are rendered on a fixed 0–100 axis and clamped visually; the label always shows the real value.

Development

npm test          # node:test smoke tests
npm run example   # writes example SVGs to examples/output/

License

MIT