Free multimodal AI available

Hi ,

Alibaba just released Qwen3.5, a multimodal AI model that handles text, images, and video.

397 billion parameters with only 17 billion activated per forward pass. Native vision-language model with 1 million token context window.

Built on hybrid architecture combining linear attention with sparse mixture-of-experts.

Available now as open-weight model.

Here’s today's client onboarding checklist prompt and why AI presentation tools still feel like AI. Then we'll look at what multimodal agents actually mean.

🔥 Prompt of the Day 🔥

Client Onboarding Checklist: Use ChatGPT or Claude

Act as an onboarding manager.

Create one client activation system for new accounts that gets campaigns live fast without missing steps.

Essential Details:

• Service Type: [WHAT YOU PROVIDE]

• Platform Focus: [WHERE YOU'LL ADVERTISE]

• Typical Launch Timeline: [DAYS TO LIVE]

• Required Assets: [WHAT CLIENT MUST PROVIDE]

• Technical Setup: [TRACKING/PIXELS/ACCOUNTS]

• Team Handoff Points: [WHO DOES WHAT]

Create one onboarding checklist including:

• Client information gathering (comprehensive form)

• Asset request list with deadlines

• Account access verification steps

• Tracking implementation sequence

• Strategy approval workflow

• Go-live quality assurance checklist

• Post-launch monitoring protocol

🤖 Tool Tuesday 🤖

AI Presentation Tools That Actually Edit

Everyone's using AI to create presentations.

Most AI-generated decks still feel like they were made by AI.

The Problem

The issue isn't the AI itself. It's how the systems are set up.

Most tools work like this:

1. You enter a prompt

2. AI generates slides

3. You're left with basic editor

4. You spend ages tweaking layouts manually

The AI and the editor are separate. They pretend to work together. They don't.

You create with AI. You edit manually. The disconnect kills efficiency.

What Actually Works

Tools where AI creates slides and edits them.

You tell the tool what you want in plain language:

"Turn slide 4 into bullet points." "Make this section more visual." "Add a chart comparing these three metrics."

No menus to search through. No drag-and-drop headaches.

Dokie AI

Does this really well.

Upload brand template. Add content. Make tweaks through conversation until it looks right.

What sets it apart:

Editing through conversation that actually works. Custom template support for brand consistency. Content-focused generation. No unnecessary fluff. In-deck image creation for unique visuals.

Why This Matters

The real shift with AI presentations isn't about making them faster.

It's about eliminating cleanup work completely.

You don't want AI to make slides you then fix manually. You want AI to make slides you can refine conversationally.

Most tools generate. Few tools edit.

The ones that do both conversationally win.

Try Dokie AI 👉here.

Did You Know?

Netflix's AI-powered recommendation system saves the company an estimated billion dollars a year by reducing subscriber churn, using machine learning to personalize content suggestions for each of its hundreds of millions of users.

🗞️ Breaking AI News 🗞️

Alibaba Launches Qwen3.5 Multimodal AI Model

Alibaba announced Qwen3.5, a native multimodal AI model that handles text, images, and video.

Open-weight model available now.

What Changed

Previous Qwen models were strong on text. Multimodal capabilities were added later.

Qwen3.5 is natively multimodal. Built from foundation to handle text, images, and video together.

Not text model with vision bolted on. Single model that understands multiple modalities natively.

The Architecture

397 billion total parameters. Only 17 billion activated per forward pass.

Sparse mixture-of-experts architecture. Most parameters stay dormant. Active subset handles each request.

This optimization means fast inference. Model size of 397B, speed of 17B.

Hybrid attention mechanism combines Gated Delta Networks (linear attention) with sparse MoE.

1 million token context window by default. Handles up to 2 hours of video input.

Performance Claims

Beats GPT-5.2, Claude 4.5 Opus, Gemini-3 Pro on multiple benchmarks.

Scores 87.8 on MMLU-Pro versus 87.4 for GPT-5.2.

88.6 on MathVision versus 83.0 for GPT-5.2.

93.1 on OCRBench versus 80.7 for GPT-5.2.

These are Alibaba's numbers. Independent verification pending.

Multimodal Capabilities

Processes images, video, and text together. Not separate processing then combining results.

Can analyze 2-hour videos within 1M token window. Understands spatial relationships in images. Handles document understanding and OCR.

Coding and agent capabilities built in. Can use tools, search web, execute code.

Agent Features

Works with coding environments like OpenClaw. Integrates with third-party agent tools.

Can perform multi-step reasoning across modalities. Visual reasoning combined with code execution.

Spatial intelligence for tasks like object counting, relative positioning, autonomous driving scene understanding.

Expanded Language Support

Grew from 119 to 201 languages and dialects.

250k vocabulary versus 150k in Qwen3. Boosts encoding/decoding efficiency 10-60% across most languages.

Strategic Context

This is China's answer to GPT-5, Claude 4.5, Gemini-3.

Alibaba competing directly with American AI companies. Open-weight model versus closed commercial models.

Released as open-weight to encourage adoption and ecosystem development.

Deployment Options

Qwen3.5-Plus: Hosted model via Alibaba Cloud Model Studio. 1M context, built-in tools, adaptive tool use.

Qwen3.5-397B-A17B: Open-weight model available for download and self-hosting.

Compatible with standard OpenAI API format. Drop-in replacement for existing integrations.

Why This Matters

China isn't just catching up to American AI. They're competing at frontier level.

Qwen3.5 claims to beat GPT-5.2, Claude 4.5, and Gemini-3 on multiple benchmarks. If true, this shifts competitive dynamics.

Open-weight release means anyone can run it. No API dependencies. No usage limits. Full control.

Multimodal from foundation means better integration across text, image, video. Not separate systems duct-taped together.

For developers, this creates alternative to American AI companies. Different regulatory environment. Different data policies. Different strategic priorities.

What This Means

If you're building AI products, Qwen3.5 offers open-weight alternative to closed commercial models.

If you're concerned about AI centralization, open-weight releases distribute capability beyond few American companies.

If you're watching AI geopolitics, this demonstrates China's frontier-level AI capabilities.

Alibaba isn't licensing American models. They're building competitive alternatives.

Over to You...

Would you actually test Qwen3.5 or just keep using what works?

Hit reply and tell me if you'd switch.

To trying new models,

Jeff J. Hunter 
Founder, AI Persona Method | TheTip.ai

NEW: “AI Money Group” to Learn to Make Money with AI