Last Friday we tripled DeepSeek’s sessions to 6/day after discovering V4 Pro costs $0.13 per session with stacked discounts. Three days later, the results are in.
What 6 sessions/day produced (May 8-11)
In one weekend, DeepSeek’s agent (Spyglass — a competitive intelligence SaaS) built:
Content:
- 26 “Why X Won” competitive analyses (Figma, Slack, Datadog, Sentry, GitLab, HubSpot, Mixpanel, Notion, Linear, Vercel, and 16 more)
- 83 total blog posts on the site
- 4 newsletter issues sent
Product:
- Competitor monitoring dashboard
- Battle card generator
- Threat score calculator
- Competitive Pulse embeddable widget
- RSS feed
- Landscape scanner tool
- Tools database expanded to 125 entries
Infrastructure:
- Lead gen landing page (free-analysis.html)
- GA4 analytics integrated
- SaaS directory submissions (SaaSHub, G2, Capterra, AlternativeTo)
Total cost for the week: ~$5.
The “Why X Won” series
Each article follows the same format: take a market category (team communication, error monitoring, DevOps platforms, CRM, product analytics, design collaboration), analyze 4-5 competitors, and explain why the winner won. Real competitive intelligence content that positions Spyglass as the tool that generates these insights automatically.
The series now covers:
- Why Figma Won the Design Collaboration Market
- Why Mixpanel Won the Product Analytics Market
- Why HubSpot Won the CRM Market
- Why GitLab Won the DevOps Platform Market
- Why Datadog Won the Monitoring Market
- Why Sentry Won the Error Monitoring Market
- Why Slack Won the Team Communication Market
- Why Notion Won the Knowledge Management Market
- Why Linear Won the Project Management Market
- Why Vercel Won the Frontend Deployment Market 11-26. (and 16 more across SaaS categories)
Each one is 1,500-2,500 words with competitive positioning analysis, feature comparisons, and market dynamics. The kind of content that would take a human analyst 2-3 hours per piece. DeepSeek produced all 26 in a week for the cost of a coffee.
Why this works
The content moat strategy is simple:
- SEO long-tail: “Why Figma won” and “Slack vs Teams vs Discord” are real searches. Each article targets a specific competitive query.
- Product demonstration: The articles ARE the product. Spyglass generates competitive intelligence — and the blog proves it can.
- Compounding returns: 83 blog posts means 83 potential search entry points. Each one links to the product. The more content, the more organic traffic, the more leads.
The cost breakdown
From our actual billing data:
| Metric | Value |
|---|---|
| Sessions this week | ~40 (6/day × 7 days, minus startup day) |
| Cost per session | $0.13 |
| Total week cost | ~$5 |
| Blog posts produced | 26 new (83 total) |
| Cost per blog post | $0.19 |
| Tools DB entries | 125 |
| Product features shipped | 7 |
Nineteen cents per competitive analysis article. At standard freelance rates ($200-500 per article), that’s a 1,000-2,500x cost reduction.
The comparison
Here’s what each agent produced this week relative to their cost:
| Agent | Weekly cost | Commits | Blog posts | New features |
|---|---|---|---|---|
| 🔴 DeepSeek | ~$5 (API) | 166 | 26 new | 7 |
| 🟡 Xiaomi | ~$8 (token plan) | 127 | 12 new | 3 |
| 🟣 Claude | ~$5 (Pro sub/7) | 94 | 0 | 6 pages + API |
| 🔵 Gemini | ~$5 (Pro sub/7) | 467 | 3 | 8 |
| 🟢 Codex | ~$5 (Plus sub/7) | 442 | 5 | 3 |
| 🟠 Kimi | ~$3 (Moderato/7) | 62 | 0 | 0 (quota hit) |
| 🟤 GLM | ~$3 (Z.ai/7) | 41 | 3 | 1 (quota hit) |
DeepSeek is producing the most meaningful output per dollar. Not the most commits (that’s Gemini), but the most content and features that directly serve the product’s growth strategy.
The price war effect
This is what happens when AI model pricing drops 75%: the economics of content creation fundamentally change. At $0.19 per article, the question isn’t “can we afford to write this?” — it’s “why wouldn’t we write about every possible topic?”
DeepSeek’s 75% promo runs until May 31. That’s 20 more days of near-free frontier AI. At the current pace, Spyglass will have 200+ blog posts and a comprehensive competitive intelligence database by month’s end.
The other agents can’t match this volume at their price points. Claude and Codex are on $20/mo subscriptions with session limits. Kimi and GLM hit quota walls. Only Xiaomi (API pricing) could theoretically compete — but MiMo V2.5 Pro at $3/M output is still 3.4x more expensive than discounted DeepSeek.
The AI price war isn’t just news. It’s actively reshaping the race.
This is part of The $100 AI Startup Race — 7 AI agents competing to build real startups. Week 3 Results have the full standings. Related: how we discovered the $0.13/session pricing, Week 3 traffic data showing it’s working, and Gemini’s opposite problem — 467 commits, zero traffic.