Cursor announced Cursor 3 on April 2, 2026—a complete product redesign around agentic coding. Code-named "Glass" during development, it's Cursor's response to Anthropic's Claude Code and OpenAI's Codex, which have captured millions of developers in recent months.

But Cursor faces a problem OpenAI and Anthropic don't: profitability pressure. While Cursor raised at a reported $50 billion valuation (nearly double its fall 2025 round), it can't afford to subsidize users the way AI labs can. OpenAI and Anthropic offer $1,000+ monthly usage for $200 subscriptions—economics that work when you're raising $100B+ but break when you need sustainable margins.

For CTOs managing developer tool budgets, this creates a strategic decision: lock into subsidized AI labs (knowing prices will rise) or bet on independent tools like Cursor (sustainable pricing but less generous limits).

For CFOs, this is the enterprise SaaS playbook in reverse: instead of vendors raising prices post-IPO, AI labs are burning capital to capture market share, then will monetize later.

What Cursor 3 (Glass) Changes

Old Cursor model: AI-powered IDE (integrated development environment)

• Developers write code in Cursor
• AI assists with autocomplete, suggestions, debugging
• Human writes code, AI helps

New Cursor 3 model: Agent-first coding

• Developers describe tasks in natural language
• AI agent completes entire features autonomously
• Human reviews code, AI writes it

Key differentiator vs Claude Code/Codex: Cursor 3 integrates agentic coding with traditional IDE in one desktop app. Developers can spin up cloud agents for feature builds, then review/refine code locally in the IDE.

Jonas Nelle (Cursor Head of Engineering): "In the last few months, our profession has completely changed. A lot of the product that got Cursor here is not as important going forward anymore."

Why Developers Are Leaving Cursor (Despite Better Product)

Developer interviews (WIRED, April 2026):

Ronald Mannak (Pico AI founder): Shifted from Cursor to Claude Code/Codex. Reason: "Whichever tool has the most generous rate limit."

Jack Crawford (mVara cofounder): Stopped using Cursor/Windsurf entirely (despite using heavily in 2025). Now uses Claude Code exclusively. Reason: "Value of the subscription."

The economics:

• Cursor pricing (post-June 2025): Usage-based (pay per token/request)
• Claude Code/Codex pricing: $200/month flat subscription
• Actual usage value: $1,000+ monthly usage (per WIRED reporting)

Why the disparity exists:

• OpenAI and Anthropic raised $100B+ → can afford to burn capital on customer acquisition
• Cursor raised ~$5B total → needs sustainable margins to survive
• AI labs view developer subscriptions as strategic loss leaders (lock in future enterprise buyers)
• Cursor views subscriptions as primary revenue stream (can't subsidize long-term)

Cursor's June 2025 pricing change:

• Removed heavily subsidized flat subscription
• Switched to usage-based pricing
• Developer backlash (TechCrunch: "Cursor apologizes for unclear pricing changes that upset users")
• Strategic rationale: "Improve margins and build a more sustainable business"

The competitive reality: Cursor can't out-subsidize AI labs with 20x more capital.

The CTO Decision Framework: Independent Tool vs AI Lab

Option 1: Cursor (Independent Tool)

Pros:

• Sustainable pricing (usage-based = predictable costs tied to actual usage)
• Product integration (agentic coding + traditional IDE in one app)
• Vendor independence (not locked into OpenAI or Anthropic ecosystem)

Cons:

• Higher cost per developer (no subsidies)
• Rate limits tighter than AI labs (can't afford to burn capital)
• Smaller model selection (relies on OpenAI/Anthropic/Google APIs + in-house Composer 2 model)

Best for:

• Teams prioritizing long-term cost predictability
• Organizations avoiding vendor lock-in
• Enterprises with compliance requirements (Cursor SOC 2 Type II certified, privacy mode)

Option 2: Claude Code / Codex (AI Lab Tools)

Pros:

• Subsidized pricing ($200/month for $1,000+ usage)
• Generous rate limits (can afford to burn capital short-term)
• Direct access to frontier models (Claude Opus, GPT-5)

Cons:

• Pricing volatility (subsidies won't last forever—Anthropic already adjusting Claude Code rate limits)
• Vendor lock-in (switching costs increase as team adopts lab-specific workflows)
• Less product integration (standalone tools, not integrated with IDE)

Best for:

• Startups maximizing runway (take advantage of subsidies while they last)
• Teams willing to accept future price increases
• Organizations already standardized on OpenAI or Anthropic

The Hybrid Strategy (Most Common in 2026)

What developers actually do (per WIRED, Builder.io, NxCode interviews):

Dual-tool approach:

• Cursor for daily coding (IDE-based work, iterative refinement, debugging)
• Claude Code for complex tasks (multi-file refactors, feature builds, architecture changes)

Rationale: Use subsidized AI lab tools for high-token tasks, use Cursor for everything else (balancing cost vs capability).

CFO implication: Budget for both tools (~$400/month per developer = Cursor usage + Claude Code subscription). Teams using hybrid approach get best economics (exploit subsidies where available, pay sustainable rates for daily work).

The Vendor Lock-In Risk: What Happens When Subsidies End?

Historical precedent: Cloud providers (AWS, Azure, GCP) offered deep discounts for enterprise migrations, then raised prices once switching costs were high.

AI labs are following same playbook:

Phase 1 (current): Subsidize developer subscriptions to capture market share

• Claude Code: $200/month for $1,000+ usage
• Codex: Similar economics
• Goal: Lock in developers before they standardize on competitors

Phase 2 (coming): Adjust rate limits, increase prices, or both

• Anthropic already adjusting Claude Code limits (March 2026 reports)
• OpenAI likely to follow (can't sustain $5:1 subsidy ratio long-term)
• Price increases will happen gradually (avoid developer backlash)

Phase 3 (future): Monetize captured user base

• Enterprises pay higher rates to retain tool their teams depend on
• Switching costs too high (re-train team, migrate workflows, adjust tooling)
• AI labs achieve sustainable margins

CTO risk assessment:

If you standardize on Claude Code/Codex today:

• 12-18 months: Enjoy subsidized pricing
• 18-24 months: Rate limits tighten, prices increase 20-30%
• 24-36 months: Pricing reaches Cursor-level economics (but you're locked in)

If you standardize on Cursor today:

• 12-18 months: Pay higher per-developer cost than AI lab tools
• 18-24 months: Cursor pricing stays stable (already at sustainable margins)
• 24-36 months: Cursor becomes cost-competitive as AI labs raise prices

CFO analysis: Subsidized tools save money short-term, cost more long-term (switching costs + eventual price increases). Independent tools cost more upfront, save money over 3-5 year TCO.

Cursor's Counter-Strategy: In-House Models

Cursor's response to subsidy war: Train proprietary models to reduce dependency on OpenAI/Anthropic APIs.

Composer 2 (launched 2026):

• Based on Moonshot AI open-source model
• Cursor did additional pretraining + post-training
• Optimized for code generation tasks
• Cost-effective to serve (no API fees to OpenAI/Anthropic)

Strategic rationale:

• Reduce variable costs (API fees to AI labs)
• Improve gross margins (in-house model inference cheaper than API calls)
• Create moat (proprietary model optimized for Cursor workflows)

Trade-off: In-house models lag frontier models (Claude Opus, GPT-5) on cutting-edge capabilities but cost less to serve.

Jonas Nelle: "People are using Composer 2 for cost-effective tasks where frontier model performance isn't required."

CTO implication: Cursor's in-house model strategy signals long-term commitment to sustainable economics, but may lag AI labs on performance for complex tasks.

Decision Criteria for 2026

Choose Cursor If:

1. ✅ You prioritize cost predictability (usage-based pricing vs volatile subscriptions)
2. ✅ You want vendor independence (not locked into OpenAI or Anthropic ecosystem)
3. ✅ You need product integration (agentic coding + IDE in one app)
4. ✅ You plan 3-5 year tool commitment (Cursor TCO wins long-term)
5. ✅ You have compliance requirements (SOC 2 Type II, privacy mode)

Choose Claude Code / Codex If:

1. ✅ You want subsidized pricing NOW (take advantage of $1,000+ monthly usage for $200)
2. ✅ You're willing to accept future price increases (lock-in trade-off)
3. ✅ You prioritize frontier model access (Claude Opus, GPT-5)
4. ✅ You're a startup maximizing runway (exploit subsidies while they last)
5. ✅ You're already standardized on OpenAI or Anthropic (ecosystem lock-in accepted)

Use Hybrid Strategy If:

1. ✅ You want best economics (exploit subsidies for high-token tasks, use Cursor for daily work)
2. ✅ You're willing to manage two tools (complexity trade-off for cost savings)
3. ✅ Your team can context-switch (different tools for different tasks)

Budget: ~$400/month per developer (Cursor usage + Claude Code subscription)

What This Means for 2026 Budgets

For CTOs:

• Evaluate 3-year TCO, not just current subscription price
• Budget for dual-tool strategy if maximizing economics (Cursor + Claude Code)
• Plan for AI lab price increases (subsidies won't last—Anthropic already adjusting)

For CFOs:

• Subsidized tools save money Year 1, cost more Years 2-3 (switching costs + price increases)
• Independent tools (Cursor) cost more upfront, save money over 3-5 year TCO
• Vendor lock-in risk: AI labs will monetize once switching costs are high

For procurement teams:

• Negotiate multi-year Cursor contracts (lock in sustainable pricing before AI labs raise prices)
• Track AI lab rate limit changes (early warning for price increases)
• Budget 20-30% annual increases for Claude Code/Codex (subsidies ending)

---

Sources:

• WIRED — Cursor Glass launch, developer interviews, subsidy economics (April 2, 2026)
• Builder.io — Claude Code vs Cursor comparison, enterprise features
• NxCode — Developer workflow analysis, hybrid strategies
• Bloomberg — Cursor $50B valuation (March 2026)
• TechCrunch — Cursor pricing change (June 2025)

---

Continue Reading

• [OpenAI Codex Pricing: $0.006/Request vs GitHub Copilot's $19/Month](/article/openai-codex-pay-as-you-go-pricing-2026)
• Scotiabank Cuts Manual Work 70% With Scotia Intelligence AI
• $40B/Year: Anthropic's Google Lock-In Reshapes AI Strategy