Blog

Microsoft Phi and Inception Mercury Are Now Available in Puter.js

We're excited to announce that Puter.js now supports Microsoft Phi and Inception Mercury, giving developers free, unlimited access to cutting-edge AI models that redefine what's possible with small language models and diffusion-based generation.

Microsoft Phi: Powerful Small Language Models

Microsoft Phi represents a paradigm shift in AI efficiency. These small language models (SLMs) deliver frontier-level capabilities while being optimized for speed, cost-effectiveness, and flexible deployment—from cloud to edge to on-device.

All Phi models feature ultra-low latency, extensive customization options, and are built following Microsoft's AI principles of accountability, transparency, fairness, and security.

// Complex reasoning with Phi 4
puter.ai.chat("Explain quantum entanglement and its implications", 
    { model: 'microsoft/phi-4' });

// Image analysis with Phi 4 Multimodal
puter.ai.chat("Describe this image", "https://example.com/image.jpg",
    { model: 'microsoft/phi-4-multimodal-instruct' });

Inception Mercury: Revolutionary Speed Through Diffusion

Mercury introduces the world's first commercial-scale diffusion large language model (dLLM), achieving speeds up to 10x faster than traditional autoregressive models.

What Makes Mercury Different:

Unlike standard language models that generate text one token at a time, Mercury uses a coarse-to-fine diffusion process that refines outputs from noise over multiple steps. This enables blazing speeds (over 1,000 tokens/second), holistic reasoning across entire outputs, continual error correction, and true parallel processing of multiple tokens at once.

// Fast text generation with Mercury
puter.ai.chat("Explain machine learning simply", 
    { model: "inception/mercury" });

// High-speed code generation with Mercury Coder
puter.ai.chat("Write a React component for a todo list", 
    { model: "inception/mercury-coder" });

Get Started Now

Just add one script tag to your HTML:

<script src="https://js.puter.com/v2/"></script>

No sign-ups, no API keys, no infrastructure setup. Start building with Microsoft Phi and Inception Mercury immediately.

Try All Models

Here's a simple demo to test all the introduced models:

<!DOCTYPE html>
<html>
<head>
    <script src="https://js.puter.com/v2/"></script>
</head>
<body>
    <h1>Microsoft Phi & Mercury Demo</h1>
    
    <h2>Microsoft Phi Models</h2>
    <button onclick="testPhi4()">Phi 4 - Complex Reasoning</button>
    <button onclick="testPhi4Plus()">Phi 4 Reasoning Plus</button>
    <button onclick="testPhi4MM()">Phi 4 Multimodal</button>
    <button onclick="testPhi4Mini()">Phi 4 Mini</button>
    <button onclick="testPhi35()">Phi 3.5</button>
    <button onclick="testPhi3()">Phi 3</button>
    
    <h2>Mercury Models (Ultra-Fast)</h2>
    <button onclick="testMercury()">Mercury</button>
    <button onclick="testMercuryCoder()">Mercury Coder</button>
    
    <div id="result" style="margin-top: 20px; padding: 15px; border: 1px solid #ccc;"></div>

    <script>
        async function testPhi4() {
            showResult("Testing Phi 4...");
            const response = await puter.ai.chat("Solve: If a train travels 60mph for 2 hours then 80mph for 1 hour, what's the average speed?", 
                { model: "microsoft/phi-4" });
            showResult(`Phi 4: ${response}`);
        }

        async function testPhi4Plus() {
            showResult("Testing Phi 4 Reasoning Plus...");
            const response = await puter.ai.chat("Plan a 3-course dinner for 6 people with $100 budget. Include shopping list.", 
                { model: "microsoft/phi-4-reasoning-plus" });
            showResult(`Phi 4 Plus: ${response}`);
        }

        async function testPhi4MM() {
            showResult("Testing Phi 4 Multimodal...");
            const response = await puter.ai.chat("Describe what you see", "https://picsum.photos/300/200", 
                { model: "microsoft/phi-4-multimodal-instruct" });
            showResult(`Phi 4 Multimodal: ${response}`);
        }

        async function testPhi35() {
            showResult("Testing Phi 3.5...");
            const response = await puter.ai.chat("Explain recursion with a simple example", 
                { model: "microsoft/phi-3.5-mini-128k-instruct" });
            showResult(`Phi 3.5: ${response}`);
        }

        async function testPhi3() {
            showResult("Testing Phi 3...");
            const response = await puter.ai.chat("Write a Python function to reverse a string", 
                { model: "microsoft/phi-3-mini-128k-instruct" });
            showResult(`Phi 3: ${response}`);
        }

        async function testMercury() {
            showResult("Testing Mercury (watch the speed!)...");
            const start = Date.now();
            const response = await puter.ai.chat("Write a creative story about AI in exactly 100 words", 
                { model: "inception/mercury" });
            const time = ((Date.now() - start) / 1000).toFixed(2);
            showResult(`Mercury (${time}s): ${response}`);
        }

        async function testMercuryCoder() {
            showResult("Testing Mercury Coder (ultra-fast coding!)...");
            const start = Date.now();
            const response = await puter.ai.chat("Create a JavaScript function to validate email addresses", 
                { model: "inception/mercury-coder" });
            const time = ((Date.now() - start) / 1000).toFixed(2);
            showResult(`Mercury Coder (${time}s): ${response}`);
        }

        function showResult(text) {
            document.getElementById('result').innerHTML = text;
        }
    </script>
</body>
</html>

This simple demo lets you test all the introduced models with one click each. Notice how Mercury models respond much faster than traditional Phi models!

Learn more:

Free, Serverless AI and Cloud

Start creating powerful web applications with Puter.js in seconds!

Get Started Now

Read the Docs Try the Playground