AI Coding Subscription Comparison 2026: Claude, Codex, and Domestic Coding Plans

2026 AI coding subscriptions have entered the 5x/20x era: Claude Pro/Max and Codex Plus/Pro are the two flagship international options, while domestic coding plans (GLM/Ark/Bailian) compete on low price and direct China access. Using official data verified 2026-08-19, this three-way comparison covers price, quota mechanics, models, ecosystem, and China accessibility - with a decision checklist you can act on directly.

Published 2026-08-21

International figures verified against official pages on 2026-08-19 (Anthropic Pricing / OpenAI Codex docs); domestic plan data from this site’s data/plans and existing roundups. The three camps are structurally very different - read the bottom line first, then the tables.

Bottom Line

  • Heavy terminal use + long code generation: Claude (Pro $20 up; Max 5x/20x) - the Opus family is the highest-rated coding model on this site (9.6/10).
  • GitHub-native workflow + multi-agent orchestration: Codex (Plus $20 up; Pro 5x/20x) - @Codex PR/issue delegation and automated code review.
  • Budget-sensitive / China-based: domestic coding plans (GLM / Ark / Bailian) - ¥50-200/month, RMB payment, direct domestic access, and coding models are now first-tier.
  • Elastic usage / cross-model needs: pay-as-you-go API or OpenRouter - don’t pay a flat subscription for occasional use.
  • Want both: entry tier on one + pay-as-you-go on the other; don’t double up on the $200 tiers.

The Three Camps at a Glance

A. Claude (Anthropic)

TierPriceNotes
Pro$20/mo ($200/yr ≈ $16.7/mo)Base usage, ~45 requests/5h
Max 5x$100/mo5x Pro usage, higher output limits, early access
Max 20x$200/mo20x usage, extreme workloads

Claude Code is included in every paid tier; the Opus family excels at long code generation and 200k context. China accessibility 2/10.

B. Codex (OpenAI / ChatGPT)

TierPriceNotes
Plus$20/moCodex App + CLI + IDE, GPT-5.6 base usage
Pro 5x$100/mo5x local-message capacity, full lineup + Codex-Spark
Pro 20x$200/mo20x capacity, unlimited Voice

GPT-5.6 in three flavors (Sol/Terra/Luna); within a tier Luna’s allowance is 20-25x Sol’s; GitHub-native (PR/issue delegation, auto code review). China accessibility 1/10.

C. Domestic Coding Plans (GLM / Ark / Bailian / MiniMax / Kimi)

Quota Mechanics Comparison

DimensionClaudeCodexDomestic Coding Plans
Billing windowrolling 5h + weekly capsrolling 5h + weekly capsvaries (monthly / daily refresh)
Quota poolshared across all surfaceslocal + cloud sharedper platform
Overageusage credits (API rates)credits (API rates)usage add-ons / downgrade
Model freedomOpus/Sonnet/Haiku familyGPT-5.6 tiers + previewdomestic models, some BYOK

Claude and Codex are near-mirrors in mechanics (which is why their prices land roughly equal); domestic plans win on “cheap + direct access” but with a narrower model lineup.

Decision Checklist

Your situationRecommendation
Heavy Claude Code in the terminal, long codeClaude Max 5x
GitHub-primary, PR/issue automationCodex Pro 5x
Multi-agent parallel + mobile controlCodex Pro 20x (unlimited Voice)
Large context + long one-shot generationClaude Max 5x
Budget-sensitive / Chinadomestic coding plan (GLM best value)
Elastic usage / cross-modelOpenRouter / API pay-as-you-go
Want bothentry tier + pay-as-you-go on the other

Further Reading

🔧 Tools in This Article

Ready to pick your tool?

Check the full overview and find your perfect AI coding setup in 30 seconds.

View Full Overview →

📖 Related Articles

📚 Related Guides