Good morning.

I gave a live talk on where agents are right now, how fast things have moved, and what it means for anyone running an online business.

I didn't sugarcoat it.

Here's the full breakdown of what I shared, expanded and cleaned up. If you were there, this is your reference doc. If you weren't, this is the session I wish someone had given me two years ago.

— Sam

Let’s get into it.

Before we get into tactics, here's what's sitting at chat.qwen.ai right now, and what comparable capability costs through other platforms.

This is a capability comparison, not a quality judgment. In some of these categories, the paid tools produce better results. In others, the gap is smaller than you'd expect. 

The point is that a year ago, getting this feature set from a single platform required multiple paid subscriptions. 

Today, a competitive version of the entire stack is available at no cost.

Go to chat.qwen.ai. Create a free account with Google or email. That's it. Desktop apps for Windows and macOS and mobile apps for iOS and Android are also available. The interface is clean and fast, similar to ChatGPT's layout.

The model powering it is Qwen3.5-Plus, which supports text, image, and video inputs. The architecture supports up to a 1 million token context window (roughly 750,000 words), though the exact context length available in the free tier may vary. Even at conservative limits, it handles far more input than most free alternatives.

Data residency. Qwen is built by Alibaba. Data is hosted on Alibaba Cloud. For research, content creation, visual generation, and general business analysis, the risk profile is comparable to any cloud-based AI tool. For genuinely sensitive data (client financials, proprietary code, trade secrets) use your judgment. If you're technical enough, the model is open-weight: download it, run it on your own hardware, and no data leaves your network.

Usage limits. The free tier is generous (more so than ChatGPT or Claude's free tiers) but there are rate limits on heavy use. Earlier Qwen products had daily request caps around 2,000. Enterprise API access through Alibaba Cloud is pay-per-token. For most operators doing research, content, and occasional image generation, the free tier covers daily workflows. If you're running automation pipelines with hundreds of API calls per day, you'll need a paid tier.

Regional variation. Some features may behave differently depending on your region. The webpage and podcast generation features from Deep Research (covered below) have been documented by VentureBeat but may not be fully available everywhere yet. Test before building workflows around specific features.

Output quality. English outputs are strong across the model's 201 supported languages, but you'll occasionally notice phrasing patterns that differ from GPT or Claude. For customer-facing copy, plan to edit. For research, analysis, and internal work, the outputs are solid without modification.

Qwen Chat has three operating modes. Knowing which one to use when is the difference between overthinking simple tasks and underthinking complex ones.

Auto Mode is the default. Qwen decides whether to think deeply, search the web, or fire off a quick response based on your prompt. It can call tools (like search, code interpreter, document analysis) adaptively. Use this for most business tasks where you want the AI to figure out the best approach on its own.

Thinking Mode is deep reasoning. Qwen shows its full chain-of-thought. You can watch it plan, reason through edge cases, and work through problems step by step. Use this for strategic analysis, complex coding, financial modeling, or any prompt where you need the AI to work through a problem rather than pattern-match a response.

Fast Mode is instant. No reasoning overhead, no thinking tokens. Use this for simple Q&A, quick content drafts, formatting tasks, and anything where speed matters more than depth.

The operator move: default to Auto. Switch to Thinking for high-stakes analysis and strategy. Switch to Fast for batch content production and simple lookups.

This is the feature that matters most for daily business use.

Qwen's Deep Research mode works like a research agent: give it a complex question, it asks clarifying questions to shape the scope, then autonomously searches the web across multiple sources, identifies discrepancies, resolves conflicts, and delivers a structured report. Outputs typically run 3,000–5,000 words with cited sources, comparisons, and actionable analysis.

This is a multi-step research execution. It plans its research path, runs queries across dozens of sources, cross-references findings, and produces something you'd normally pay a research assistant or junior analyst to compile.

A competitor analysis prompt returns a structured report with current pricing tables, feature comparisons, deliverability benchmarks, integration counts, and strategic gaps, complete with source URLs:

A market research prompt returns the kind of landscape analysis that would take a human researcher a full day or more to assemble: 

The practical advantage: most operators use AI for quick Q&A. They're not running structured research because the free tiers of other platforms don't support it well, and the paid tiers have tight usage limits. 

Qwen's Deep Research offers enough free usage to build a weekly research habit (competitor monitoring, market scanning, content research) without hitting a paywall on normal usage.

VentureBeat recently covered an expansion to Qwen Deep Research. After generating a report, users can reportedly convert it into a live hosted webpage (with inline graphics, clean typography, and a shareable URL) and a multi-speaker podcast (using Qwen's text-to-speech model with multiple narrators). If this works in your region, one research prompt produces a report, a hosted webpage, and an audio version. Test before building a workflow around it.

The meaningful differentiator here: accurate text rendering inside generated images.

This sounds minor until you've tried generating an infographic, social media graphic, or promotional image in Midjourney or DALL-E. The text comes out garbled or misspelled. You end up in Canva anyway, layering text on top.

Nano Banana 2 is, of course, the gold standard for this but it can fail more often than not.

Qwen-Image 2.0 generates at native 2K resolution (2048×2048) and supports prompts up to 1,000 tokens, so you can describe detailed layouts. It currently ranks first on AI Arena (a blind human evaluation platform) for both text-to-image generation and image editing.

What this means in practice: 

You can generate in one shot infographics with properly formatted text, presentation slides with accurate titles and layout structure, social media graphics with readable headlines and CTAs, and multi-panel visual content with consistent typography.

This doesn't replace a senior designer for brand-critical work. It replaces the routine visual content that operators currently handle through Canva Pro subscriptions, freelance designers, or creative agencies.

Qwen Chat includes text-to-video generation — describe a scene and get a short video (up to 10 seconds) with synchronized audio. The quality is early-stage, comparable to early Runway outputs. 

For social media clips, product teasers, ad concepts, and internal presentations, it works. For anything polished, you'll need dedicated tools.

Describe a webpage in plain English. Get a complete, functional HTML/CSS/JS page you can preview live and download.

A prompt like this one produces a working, responsive page:

"Build a landing page for an AI consulting agency. Hero section with headline, three-column features section, client testimonials, pricing table with three tiers, and a contact form. Modern dark theme, clean typography, subtle scroll animations."

The workflow is simple: describe your page in Web Dev mode. Preview it live. Download the code. Deploy on Netlify or Vercel. Total time from concept to live page: about 15 minutes. The code is clean enough to hand to a developer for refinement if you need to customize beyond what the initial generation produces.

Qwen3.5's multimodal capabilities combined with its context window is great for a use case like: 

Feeding substantial business context into a single prompt and getting strategic analysis back.

You can upload screenshots of analytics dashboards (Google Analytics, Shopify, Stripe), financial reports, business plans, customer data spreadsheets, competitor website screenshots, and ad creative images — and process far more pages in a single prompt than most free alternatives support.

HubSpot's SVP of Marketing tested an earlier Qwen model by feeding it business metrics and asking it to identify trends, prioritize metrics, build a forecast, and create a strategy. 

Qwen identified key growth levers, flagged core bottlenecks, and created the forecast — performing comparably to ChatGPT's paid tier on a multi-step strategic analysis. 

Qwen excels when the model needs to work through five or more steps to complete an objective. That's exactly where simpler tools start to fall apart. Qwen3.5 is a significant upgrade from the version he tested.

The practical application: take screenshots of a client's analytics dashboards, upload them, and prompt "Analyze these dashboards. Identify the three most concerning trends, the two biggest opportunities, and give me a prioritized list of five actions to improve performance over the next 30 days. Be specific about which pages, channels, and metrics to focus on." Run that across five client dashboards in 20 minutes. That's a meaningful compression of analysis time.

Everything above is the current free chat interface. The Qwen3.5 model release signals something larger that operators building AI-native systems should be tracking.

Qwen3.5 was explicitly designed as infrastructure for autonomous agents — AI systems that plan multi-step workflows, reason through complex decisions, use tools, and execute tasks independently. The model posted strong results on BrowseComp (a benchmark testing autonomous web browsing and task completion), outperforming Gemini 3 Pro and performing competitively with Claude Opus 4.5.

Here's the capability most coverage has missed: Qwen3.5 can act as a visual agent for productivity automation. It autonomously interacts with smartphone and desktop interfaces — navigating apps, clicking buttons, filling spreadsheets, handling multi-step workflows.

In demos, it filled missing data in Excel, navigated across multiple mobile apps following natural-language instructions, and completed long-horizon desktop workflows for office automation. It scored 62.2 on OSWorld (a verified desktop automation benchmark) and 66.8 on AndroidWorld (mobile automation).

This is early. But the trajectory is clear: the models are moving from generating text to operating software. For operators who are already thinking about autonomous workflows, this is the section of the Qwen release that deserves the most attention.

With expanded context capacity, Qwen3.5 can process long-form video — the architecture supports up to two hours at full context length. Official demos show it turning hand-drawn UI sketches into clean frontend code, reverse-engineering game logic from gameplay footage, and condensing long videos into structured web pages. For course creators analyzing competitor content or operators repurposing video into documentation, this capability didn't exist six months ago.

Released under Apache 2.0, anyone can download, customize, and deploy Qwen3.5. No API dependency. No vendor lock-in. No usage caps. If you're working with a developer or technical co-founder, you can train this model on your specific business data and build proprietary tools that competitors can't replicate. Qwen3.5 uses a Mixture of Experts architecture — 397 billion total parameters, 17 billion active per task — which means running agents and processing pipelines at a fraction of what comparable systems cost on GPT or Claude APIs.

L'Oréal China partnered with Alibaba Cloud to deploy Qwen across multiple business functions. They built an agentic AI application powered by a domain knowledge base that delivers intelligent Q&A for frontline skincare practitioners, improving communication efficiency with consumers. Their IT organization uses Alibaba Cloud's AI coding assistant (powered by Qwen) to accelerate software development. If a $40B beauty conglomerate is running Qwen at enterprise scale for customer-facing agents and internal development, the model's quality is proven beyond benchmarks.

Qwen models have been deployed by over 90,000 enterprises through Alibaba Cloud's Model Studio. Over 2.2 million corporate users access Qwen-powered services through DingTalk (Alibaba's collaboration platform). The open-source models have seen over 7 million downloads on HuggingFace and GitHub.

Greyhound Research's chief analyst noted that Qwen3.5 functions as a "workflow-capable system" — when reasoning, multimodal understanding, and tool use combine, it stops behaving like a conversational assistant and starts behaving like an execution layer.

Gartner's senior director analyst confirmed strong multimodal capabilities with extensive model selection and open model options for customization, while noting that global adoption outside China is still developing.

Here are ready-to-use workflows across business types — built around what Qwen Chat can do right now, for free.

Weekly client reporting. Screenshot each client's key dashboards (ad platforms, analytics, CRM). Upload to Qwen Chat: "Analyze these dashboards for [client name]. Summarize performance trends, flag anything concerning, identify top opportunities, and draft a 500-word client update email with three recommended next steps." Edit for tone and specifics. Send.

Lead magnet creation. Use Deep Research to identify top pain points in your target vertical. Draft a comprehensive guide addressing those pain points. Use Qwen-Image 2.0 to generate the cover graphic and internal visuals with accurate text. Use Web Dev mode to build the landing page. Complete lead gen asset, start to finish, in a few hours.

Competitive intelligence. Set up a weekly Deep Research prompt: "Research new developments, funding rounds, product launches, and pricing changes from [competitor 1], [competitor 2], and [competitor 3] in the past 7 days. Summarize findings and flag anything requiring strategic response." Share as a recurring weekly brief with your team.

Feature page prototyping. Use Web Dev mode to rapidly prototype feature pages, comparison pages, and campaign landing pages. Test messaging before investing in Webflow or a developer. Idea to live prototype in about 15 minutes.

Content research. Use Deep Research to analyze what's currently being taught in your niche: "Research the top 10 online courses in [niche]. For each, identify curriculum structure, pricing, student reviews, and completion rates where available. Identify the three biggest gaps a new course could fill."

Sales page creation. Draft your copy in Qwen Chat using existing course details. Use Web Dev mode to build the full sales page. Use Qwen-Image 2.0 for hero images, testimonial graphics, and module preview visuals — all with readable text. Deploy directly.

Content research pipeline. Use Deep Research weekly: "Research the most important developments in [your topic] from the past 7 days. For each, explain what happened, why it matters, and what the implications are for [your audience]. Organize by importance."

Visual content. Generate custom header images, infographics, and section dividers for each issue using Qwen-Image 2.0 — with your newsletter's name, issue number, and section headers rendered accurately in the images.

Product listing optimization. Upload screenshots of your product pages. "Audit these product listings for conversion optimization. Analyze the titles, descriptions, image quality, pricing presentation, and review placement. Rewrite the titles and descriptions to improve click-through and conversion rates."

Ad creative generation. Use Qwen-Image 2.0 to generate multiple ad creative variations with actual text rendered in the images. Generate five variations with different headlines and layouts. A/B test the winners.

Here's what matters beyond any single tool or prompt.

Frontier-class AI capability is commoditizing faster than anyone predicted. Qwen3.5 is one data point, but it's a loud one. A year ago, this level of performance was locked behind expensive proprietary APIs. Today it's open-weight and available on a free platform.

Three implications worth thinking through:

Your AI tool subscriptions face a new burden of proof. When a free tier covers a substantial portion of what you're paying $20-200/month for, the paid tools need to justify themselves on the margin where they're measurably better. Some will. Some are coasting on switching costs and habit. Audit your current stack. For each tool, run the same prompt in Qwen Chat and your paid tool. Compare the outputs honestly. You may find that some subscriptions are earning their fee and others aren't — and now you have the evidence to make that call.

The competitive advantage shifts from access to deployment. When intelligence is widely available, having AI tools isn't an edge. Knowing how to deploy them is. The operators who benefit from this shift aren't the ones who find the cheapest tools. They're the ones who build the best systems — the research pipelines, content workflows, analysis frameworks, and agent architectures that turn raw capability into business output. That's a skill gap, not a budget gap. And skill gaps favor the people who build systems while everyone else is still evaluating which chatbot to subscribe to.

The minimum viable AI stack just got lighter. A solo founder can now run research, content, design, video, and web development operations with zero marginal cost for the AI layer. That doesn't mean you should replace everything with free tools. It means the baseline has shifted. Start with the free tier. Add paid tools only when they demonstrably outperform on specific tasks that matter to your business. The paid tools aren't going away — but they need to earn their place against a baseline that's getting more capable every quarter.

Alibaba didn't release Qwen3.5 to be generous.

They released it to win the infrastructure layer: to become the foundation that millions of businesses build on top of.

The free access is a growth strategy, not a gift.

But that doesn't change what it means for you.

Today, you can open a browser tab and access an AI platform with deep research, image generation with accurate text rendering, website creation from plain English, multimodal document analysis, and a context window larger than most free alternatives. It's competitive with models that cost significantly more. Over 90,000 enterprises are running on it. And the free tier covers most daily operator workflows.

Is it perfect? No. You'll hit some limits. Some features may not be fully rolled out in your region. The phrasing occasionally differs from what Western-trained models produce. For customer-facing copy, you'll want to edit.

Here's your action for this week: go to chat.qwen.ai.

Run one Deep Research report on your biggest competitor.

Generate one piece of visual content with Qwen-Image 2.0 and see whether the text renders correctly for your use case.

Build one landing page using Web Dev mode.

Compare the quality to what you're currently paying for.

Let the results tell you what's worth keeping and what isn't.

Until next time,
Sam Woods
The Editor

.