The New Codex Subscription Landscape: Free, Go, Plus, Pro and Enterprise Compared

pricingsubscription-tierspro-plancodex-sparkenterprisecost-managementclaude-maxcompetitive-analysis

The New Codex Subscription Landscape: Free, Go, Plus, Pro and Enterprise Compared

On 9 April 2026, OpenAI announced a new $100/month ChatGPT Pro tier that slots between the $20 Plus plan and the former $200 Pro plan — now rebranded as Pro 20x 1. Combined with the $8 Go tier, the pay-as-you-go Codex-only seats launched on 3 April 2, and the model deprecation wave that began on 7 April 3, the Codex subscription landscape has been comprehensively restructured in under a week. This article maps every tier, explains what each means for Codex CLI practitioners, and compares the result with Anthropic’s Claude Max pricing.

Why the Restructure Matters

Codex now has over 3 million weekly active users — a fivefold increase in three months, with 70% month-over-month growth 4. That explosive adoption exposed a pricing gap: developers who exceeded Plus limits had no option short of the $200 Pro plan. The new $100 tier directly addresses this, whilst the pay-as-you-go Codex-only seats give enterprise teams granular cost control for CI/CD and multi-agent workloads 2.

The Complete Tier Map

graph TD
    A[Free<br/>$0/month] --> B[Go<br/>$8/month]
    B --> C[Plus<br/>$20/month]
    C --> D[Pro 5x<br/>$100/month]
    D --> E[Pro 20x<br/>$200/month]

    F[Business<br/>$20/seat/month] --> G[Codex-Only Seat<br/>Pay-as-you-go]
    H[Enterprise<br/>Contact Sales] --> G
    I[API Key<br/>Token-based] --> J[No subscription<br/>required]

    style A fill:#f0f0f0
    style B fill:#e8f4e8
    style C fill:#d0e8ff
    style D fill:#ffe0b0
    style E fill:#ffd0d0
    style F fill:#e8e0f8
    style G fill:#fff0d0
    style H fill:#e8e0f8
    style I fill:#f0f0f0
    style J fill:#f0f0f0

Consumer and Individual Tiers

Tier Price Codex Local Messages (5 hr) Cloud Tasks Spark Access Key Use Case
Free $0 Limited — — Quick exploration
Go $8/mo Light — — Lightweight coding tasks
Plus $20/mo 20–100 (GPT-5.4) / 60–350 (mini) / 30–150 (5.3-Codex) Limited — Daily development
Pro 5x $100/mo 200–1,000 (GPT-5.4) / 600–3,500 (mini) / 300–1,500 (5.3-Codex) Available ✅ Research preview Professional daily use
Pro 20x $200/mo 400–2,000 (GPT-5.4) / 1,200–7,000 (mini) / 600–3,000 (5.3-Codex) Available ✅ Research preview Intensive parallel workflows

All usage limits operate on a rolling 5-hour window rather than a daily cap 5.

Team and Enterprise Tiers

Tier Price Differentiator
Business $20/seat/mo (down from $25) Standard seats with Codex usage cap; optional Codex-only seats
Codex-Only Seat Pay-as-you-go (token-based) No rate limits; billed on consumption; up to $500 promo credits per member 2
Enterprise Contact sales SAML SSO, SCIM, EKM, RBAC, audit logs, data retention, credit pools
Edu Contact sales Same enterprise security, education pricing

API Key Access

Developers can bypass subscription tiers entirely by authenticating with an API key 5. This provides:

  • Token-based billing at published API rates
  • No cloud features (cloud tasks require ChatGPT authentication)
  • Access to any API-supported model, including custom providers via [model_providers] in config.toml
  • No rate limit resets tied to subscription events

The Pro 5x Sweet Spot

The $100 Pro tier is the headline change. For Codex CLI users, the practical impact is substantial:

5× the message throughput. Where Plus gives you 20–100 GPT-5.4 local messages per 5-hour window, Pro 5x gives you 200–1,000 5. For a typical subagent workflow running an orchestrator plus three workers, that is the difference between hitting the wall mid-afternoon and running comfortably through a full working day.

GPT-5.3-Codex-Spark access. Spark — the Cerebras-powered model running at 1,000+ tokens per second — remains exclusive to Pro subscribers 6. For interactive refinement workflows where near-instant feedback changes how you work, this alone may justify the upgrade from Plus.

Promotional 10× boost through May 2026. Until 31 May, Pro 5x subscribers receive a temporary 2× multiplier on top of the standard 5×, effectively giving 10× Plus usage 1. The shown limits on the pricing page already include this boost — they will halve on 1 June.

Configuring Spark in Your Profile

# ~/.codex/config.toml

[profiles.spark]
model = "gpt-5.3-codex-spark"
model_reasoning_effort = "high"

[profiles.spark.model_providers.openai]
# Spark requires ChatGPT authentication, not API key

Then invoke with:

codex --profile spark "refactor the auth module to use PKCE"

Token Economics: Credits vs API Rates

Understanding the two billing models is critical for cost planning.

Credit-Based Billing (Plus / Pro)

Credits are consumed per task, with costs varying by model and task type 5:

Task Type GPT-5.4 GPT-5.3-Codex GPT-5.4-mini
Local task ~7 credits ~5 credits ~2 credits
Cloud task ~34 credits ~25 credits —
Code review ~34 credits ~25 credits —

Fast mode doubles credit consumption. Credits enable continued use beyond included limits.

Token-Based Billing (Business / Codex-Only Seats)

For pay-as-you-go seats, costs are calculated per million tokens 2:

Model Input (credits/1M) Cached Input Output
GPT-5.4 62.50 6.25 375.00
GPT-5.4-mini 18.75 1.875 113.00
GPT-5.3-Codex 43.75 4.375 350.00

The cached input rate — roughly 10× cheaper than uncached — makes prompt caching and session resumption significant cost levers 5.

The Competitive Mirror: Claude Max

Anthropic’s Claude pricing now mirrors OpenAI’s structure almost exactly 7:

  OpenAI Codex Anthropic Claude Code
Base tier Plus $20/mo Pro $20/mo
Mid tier Pro 5x $100/mo Max 5x $100/mo
Top tier Pro 20x $200/mo Max 20x $200/mo
Usage model 5-hour rolling window ~5-hour rolling window
Pay-as-you-go Codex-only seats (token-based) API key only

The structural similarity is not coincidental. Both companies are converging on the same insight: professional developers will pay $100/month for 5× throughput, but the $20-to-$200 jump was losing them to the competitor 8.

Key differentiators remain:

  • Codex offers pay-as-you-go enterprise seats with no rate limits — Claude has no equivalent 2
  • Codex has Spark for near-instant feedback loops — Claude has no speed-optimised model tier 6
  • Claude Code scores higher on Terminal-Bench (39th place notwithstanding, Claude’s Opus 4.6 model is generally rated #1 on multi-file refactoring tasks) 8
  • Claude Max includes priority access to new models and features like voice mode 7

Decision Framework: Which Tier for Which Workflow

flowchart TD
    A[How do you use Codex?] --> B{Occasional use?}
    B -->|Yes| C[Plus $20/mo]
    B -->|No| D{Multi-agent or<br/>heavy daily use?}
    D -->|Light-medium| E{Need Spark<br/>for real-time?}
    E -->|No| C
    E -->|Yes| F[Pro 5x $100/mo]
    D -->|Heavy| G{Parallel agents<br/>or cloud tasks?}
    G -->|Yes, moderate| F
    G -->|Yes, intensive| H[Pro 20x $200/mo]
    A --> I{Team or enterprise?}
    I -->|Yes| J{Predictable usage?}
    J -->|Yes| K[Business $20/seat]
    J -->|Variable or CI/CD| L[Codex-only seats<br/>Pay-as-you-go]
    A --> M{Want full control<br/>or custom models?}
    M -->|Yes| N[API Key<br/>Token-based]

Practical Recommendations

Solo developer, daily use: Start with Plus. If you consistently hit the 5-hour ceiling before lunch, upgrade to Pro 5x. The promotional 10× boost through May gives you a month to evaluate whether the extra headroom is worth it.

Subagent-heavy workflows: Pro 5x is the minimum. A typical orchestrator + 3 worker pattern can consume 15–30 messages per turn cycle. At Plus limits (20–100 messages per window), you get 1–6 full cycles. At Pro 5x (200–1,000), you get 13–66.

CI/CD pipelines: Use Codex-only seats with token-based billing. Rate limits are absent, and you pay only for what codex exec actually consumes 2. Set max_tokens_per_session in your CI profile to cap runaway costs:

# .codex/config.toml (project-level)

[profiles.ci]
model = "gpt-5.4-mini"
model_reasoning_effort = "medium"
approval_policy = "full-auto"
sandbox_mode = "read-only"

Enterprise teams (50+ developers): The Business seat price drop to $20/month makes standard seats cheaper 2, whilst Codex-only seats let you allocate budget to the developers and pipelines that need it. Enterprise credit pools enable department-level budgeting without per-developer licensing friction.

The Model Availability Shift

The 7 April model picker update removed six models from ChatGPT-authenticated sessions 3:

  • gpt-5.2-codex — removed from picker, full removal 14 April
  • gpt-5.1-codex-mini, gpt-5.1-codex-max, gpt-5.1-codex — deprecated 1 April, removed from picker
  • gpt-5.1, gpt-5 — removed from picker

The remaining models for ChatGPT sign-in are:

  • gpt-5.4 (recommended default)
  • gpt-5.4-mini (subagent workhorse)
  • gpt-5.3-codex (coding specialist)
  • gpt-5.2 (legacy, still available)
  • gpt-5.3-codex-spark (Pro only, research preview)

API key users can still access other API-supported models and configure custom providers 3. If your config.toml references any deprecated model, Codex will fall back to the default — update your profiles now to avoid surprises:

# Check all your config files for deprecated models
grep -r "gpt-5.1\|gpt-5.0\|gpt-5\"" ~/.codex/ .codex/ 2>/dev/null

What This Means for the Ecosystem

The restructured pricing tells a clear story about where OpenAI sees Codex heading:

  1. Codex is the growth engine. The 3 million weekly users and 70% month-over-month growth justify the entire pricing restructure 4. The $100 tier exists because Codex created demand for it.

  2. Enterprise is pay-as-you-go. Fixed per-seat licensing is giving way to consumption-based billing — the same model that won for cloud infrastructure. The $500 promotional credits per Codex-only seat lower the barrier for enterprise evaluation 2.

  3. Speed is a premium feature. Spark access gated behind Pro signals that OpenAI views inference speed as a differentiation axis, not just model quality 6.

  4. The Claude Code arms race is symmetrical. Both companies now offer identical tier structures at identical price points. The battleground has shifted from pricing to capability — which matters more for your workflow: Codex’s Spark speed and cloud task architecture, or Claude Code’s multi-file refactoring strength and extended thinking 8.

Citations

  1. TechCrunch, “ChatGPT finally offers $100/month Pro plan,” 9 April 2026. https://techcrunch.com/2026/04/09/chatgpt-pro-plan-100-month-codex/ ↩ ↩2

  2. OpenAI, “Codex now offers pay-as-you-go pricing for teams,” 3 April 2026. https://openai.com/index/codex-flexible-pricing-for-teams/ ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8

  3. OpenAI Developer Changelog, “Codex Model Availability Update,” 7 April 2026. https://developers.openai.com/codex/changelog ↩ ↩2 ↩3

  4. BusinessToday, “OpenAI Codex celebrates 3 million weekly users, CEO Sam Altman resets usage limits,” 8 April 2026. https://www.businesstoday.in/technology/story/openai-codex-celebrates-3-million-weekly-users-ceo-sam-altman-resets-usage-limits-524717-2026-04-08 ↩ ↩2

  5. OpenAI Developer Docs, “Codex Pricing,” April 2026. https://developers.openai.com/codex/pricing ↩ ↩2 ↩3 ↩4 ↩5

  6. OpenAI Developer Docs, “Codex Models,” April 2026. https://developers.openai.com/codex/models ↩ ↩2 ↩3

  7. IntuitionLabs, “Claude Max Plan Explained: Pricing, Limits & Features,” 2026. https://intuitionlabs.ai/articles/claude-max-plan-pricing-usage-limits ↩ ↩2

  8. Dataconomy, “OpenAI Launches New $100-a-month Pro Plan For ChatGPT,” 10 April 2026. https://dataconomy.com/2026/04/10/openai-launches-new-100-a-month-pro-plan-for-chatgpt/ ↩ ↩2 ↩3