Tell the advisor your workload — it checks every plan's included consumption in that plan's own unit (Claude weekly hours, Cursor monthly API-dollar pools, ChatGPT messages per session, metered dollars on OpenRouter and OpenCode Go), shows the smartest model each plan unlocks with measured intelligence scores, and recommends the cheapest plan that fits.
Set what you can spend per month first — plans above your budget dim and leave the recommendations. Then tune volume, task size, rhythm, and the intelligence floor below.
$10/mo$100/mo$200/mo
Plans above your budget stay visible but dimmed, and leave the recommendations. Per-seat plans compare on per-seat fee.
Claude plans lock you to Claude Code / Cowork / Agent SDK — subscription auth in other harnesses risks a ban. ChatGPT, OpenRouter, and OpenCode Go work anywhere.
50 (light)1000 (daily agent use)2000 (heavy)
Plans whose smartest model scores below this are excluded.
What does a task look like?
Usage rhythm
Best fit
Claude Pro
Cheapest plan that fits with comfortable headroom.
Cheapest that fits
Claude Pro
Lowest fee among fitting plans.
Most headroom
ChatGPT Pro 100
Largest capacity margin for growth and bursts.
Most portable fit
OpenCode Go $20
Best fitting plan that works in third-party tools.
Claude Pro
Claude · $20/mo ($17/mo annual)
Fits
Your need: 35 hrs/wkIncluded: 40–80 Sonnet-equiv hrs/wk · official
Headroom bar: your need vs included capacity.
Smartest model
Opus 5 · intel 50.7
measured score · —
Windows
At least 5x Free usage per 5-hour session
40-80 Sonnet-equiv hrs/wk (official); separate Opus weekly cap
Tool integration
Harness lockedClaude Code, Claude Cowork, Claude Agent SDK (official Anthropic surfaces only)Anthropic ToS bans subscription auth in unapproved third-party harnesses and CLIs — accounts get rate-limited or banned. Use API keys for other tools.
$20/mo ≈ $0.067/task at 300 tasks/mo · ≈$0.16/task API value
Claude Max 5x
Claude · $100/mo
Fits
Your need: 35 hrs/wkIncluded: 140–280 Sonnet-equiv hrs/wk · official
Headroom bar: your need vs included capacity.
Smartest model
Fable 5 1 · intel 53.4
measured score · —
Windows
5x Pro usage per 5-hour session
140-280 Sonnet + 15-35 Opus hrs/wk (official); priority access
Tool integration
Harness lockedClaude Code, Claude Cowork, Claude Agent SDK (official Anthropic surfaces only)Anthropic ToS bans subscription auth in unapproved third-party harnesses and CLIs — accounts get rate-limited or banned. Use API keys for other tools.
$100/mo ≈ $0.333/task at 300 tasks/mo · ≈$0.32/task API value
ChatGPT Plus
ChatGPT · $20/mo
Tight
Your need: 45 msgs/sessionIncluded: 10–100 Sol local msgs/5h · official
Headroom bar: your need vs included capacity.
Smartest model
5 6 Sol · intel 47.1
measured score · Codex expanded
Windows
10-100 Sol msgs/5h; Terra 25-200, Luna 250-2000 (official)
Weekly limits may also apply (amounts unpublished)
Tool integration
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
$20/mo ≈ $0.067/task at 300 tasks/mo · ≈$0.04/task API value
ChatGPT Pro 100
ChatGPT · $100/mo
Fits
Your need: 45 msgs/sessionIncluded: 50–500 Sol local msgs/5h · official
Headroom bar: your need vs included capacity.
Smartest model
6 Astra · intel 52.8
measured score · Codex maximum
Windows
50-500 Sol msgs/5h (5x Plus, official)
Weekly limits may also apply (amounts unpublished)
Tool integration
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
$100/mo ≈ $0.333/task at 300 tasks/mo · ≈$0.36/task API value
ChatGPT Business
ChatGPT · $25/mo ($20/mo annual)
Tight
Your need: 45 msgs/sessionIncluded: 10–100 Sol local msgs/5h (Standard) · official
Headroom bar: your need vs included capacity.
Smartest model
5 6 Sol · intel 47.1
measured score · Codex expanded
Windows
Plus-level Codex allowance on Standard; higher tiers ~Pro 5x (official)
Admin controls; workspace credits can extend access
Tool integration
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
$25/mo ≈ $0.083/task at 300 tasks/mo · ≈$0.04/task API value
Per seat.
Cursor Pro
Cursor · $20/mo
Over limit
Your need: $69–$238Included: 20–20 API-value $/mo included · community
Headroom bar: your need vs included capacity.
Smartest model
Opus 5 · intel 50.7
measured score · Cursor Agent + Tab
Windows
No 5h window; monthly pools (Cursor Models + Other Models)
Monthly included usage; on-demand top-up at API rates
Tool integration
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's ownSubscription can't drive other agent harnesses; for other tools you'd pay per-token elsewhere.
$20/mo ≈ $0.067/task at 300 tasks/mo · API-value burn $0.23–$0.79/task across models
Cursor Pro Plus
Cursor · $60/mo
Over limit
Your need: $69–$238Included: 60–60 API-value $/mo included · community
Headroom bar: your need vs included capacity.
Smartest model
Opus 5 · intel 50.7
measured score · Cursor Agent + Tab
Windows
No 5h window; monthly pools (Cursor Models + Other Models)
Monthly included usage; on-demand top-up at API rates
Tool integration
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's ownSubscription can't drive other agent harnesses; for other tools you'd pay per-token elsewhere.
$60/mo ≈ $0.200/task at 300 tasks/mo · API-value burn $0.23–$0.79/task across models
Cursor Teams Standard
Cursor · $40/mo ($40/mo annual)
Over limit
Your need: $69–$238Included: 40–40 API-value $/mo/seat included · community
Headroom bar: your need vs included capacity.
Smartest model
Opus 5 · intel 50.7
measured score · Cursor Agent + Tab
Windows
Per-seat monthly pool; centralized billing and admin
Team marketplace, Bugbot reviews, usage analytics, SSO
Tool integration
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's ownSubscription can't drive other agent harnesses; for other tools you'd pay per-token elsewhere.
$40/mo ≈ $0.133/task at 300 tasks/mo · API-value burn $0.23–$0.79/task across models
Per seat.
OpenRouter $20 budget
OpenRouter · $20/mo
Tight
Your need: $11–$238Included: 20–20 API-value $/mo budget · derived
Headroom bar: your need vs included capacity.
Smartest model
Fable 5 1 · intel 53.4
measured score · Pay per token
Windows
No windows; pure metered
Spend-capped only; pick any model per task
Tool integration
Works in any toolAny OpenAI-compatible tool via API key (agents, CLIs, IDEs, scripts)
$20/mo ≈ $0.067/task at 300 tasks/mo · API-value burn $0.04–$0.79/task across models
OpenRouter $50 budget
OpenRouter · $50/mo
Tight
Your need: $11–$238Included: 50–50 API-value $/mo budget · derived
Headroom bar: your need vs included capacity.
Smartest model
Fable 5 1 · intel 53.4
measured score · Pay per token
Windows
No windows; pure metered
Spend-capped only; pick any model per task
Tool integration
Works in any toolAny OpenAI-compatible tool via API key (agents, CLIs, IDEs, scripts)
$50/mo ≈ $0.167/task at 300 tasks/mo · API-value burn $0.04–$0.79/task across models
OpenRouter $100 budget
OpenRouter · $100/mo
Tight
Your need: $11–$238Included: 100–100 API-value $/mo budget · derived
Headroom bar: your need vs included capacity.
Smartest model
Fable 5 1 · intel 53.4
measured score · Pay per token
Windows
No windows; pure metered
Spend-capped only; pick any model per task
Tool integration
Works in any toolAny OpenAI-compatible tool via API key (agents, CLIs, IDEs, scripts)
$100/mo ≈ $0.333/task at 300 tasks/mo · API-value burn $0.04–$0.79/task across models
OpenCode Go $20
OpenCode Go · $20/mo
Tight
Your need: $5.5–$69Included: 10–10 Zen credits $/mo usable · derived
Headroom bar: your need vs included capacity.
Smartest model
4 6 · intel 44.4
measured score · Zen credits
Windows
No windows; credit-metered
Credit-capped only; open coding models via Zen
Tool integration
Works in any toolAny tool via OpenCode Zen OpenAI-compatible endpoints (chat/completions, responses, messages)
$20/mo ≈ $0.067/task at 300 tasks/mo · API-value burn $0.02–$0.23/task across models
OpenCode Go $50
OpenCode Go · $50/mo
Tight
Your need: $5.5–$69Included: 40–40 Zen credits $/mo usable · derived
Headroom bar: your need vs included capacity.
Smartest model
4 6 · intel 44.4
measured score · Zen credits
Windows
No windows; credit-metered
Credit-capped only; open coding models via Zen
Tool integration
Works in any toolAny tool via OpenCode Zen OpenAI-compatible endpoints (chat/completions, responses, messages)
$50/mo ≈ $0.167/task at 300 tasks/mo · API-value burn $0.02–$0.23/task across models
OpenCode Go $100
OpenCode Go · $100/mo
Fits
Your need: $5.5–$69Included: 90–90 Zen credits $/mo usable · derived
Headroom bar: your need vs included capacity.
Smartest model
4 6 · intel 44.4
measured score · Zen credits
Windows
No windows; credit-metered
Credit-capped only; open coding models via Zen
Tool integration
Works in any toolAny tool via OpenCode Zen OpenAI-compatible endpoints (chat/completions, responses, messages)
$100/mo ≈ $0.333/task at 300 tasks/mo · API-value burn $0.02–$0.23/task across models
Cursor Teams Premium
Cursor · $120/mo ($120/mo annual)
Tight
Over your $100 budget
Your need: $69–$238Included: 120–120 API-value $/mo/seat included · community
Headroom bar: your need vs included capacity.
Smartest model
Opus 5 · intel 50.7
measured score · Cursor Agent + Tab
Windows
Per-seat monthly pool; centralized billing and admin
Team marketplace, Bugbot reviews, usage analytics, SSO
Tool integration
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's ownSubscription can't drive other agent harnesses; for other tools you'd pay per-token elsewhere.
$120/mo ≈ $0.400/task at 300 tasks/mo · API-value burn $0.23–$0.79/task across models
Per seat.
ChatGPT Pro 20x
ChatGPT · $200/mo
Fits
Over your $100 budget
Your need: 45 msgs/sessionIncluded: 200–2,000 Sol local msgs/5h · official
Headroom bar: your need vs included capacity.
Smartest model
6 Astra · intel 52.8
measured score · Codex maximum
Windows
200-2000 Sol msgs/5h (20x Plus, official)
Weekly limits may also apply (amounts unpublished)
Tool integration
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
$200/mo ≈ $0.667/task at 300 tasks/mo · ≈$0.36/task API value
Claude Max 20x
Claude · $200/mo
Fits
Over your $100 budget
Your need: 35 hrs/wkIncluded: 560–1,120 Sonnet-equiv hrs/wk · derived
Headroom bar: your need vs included capacity.
Smartest model
Fable 5 1 · intel 53.4
measured score · —
Windows
20x Pro usage per 5-hour session
~4x Max 5x by tier ratio (derived); priority access
Tool integration
Harness lockedClaude Code, Claude Cowork, Claude Agent SDK (official Anthropic surfaces only)Anthropic ToS bans subscription auth in unapproved third-party harnesses and CLIs — accounts get rate-limited or banned. Use API keys for other tools.
$200/mo ≈ $0.667/task at 300 tasks/mo · ≈$0.32/task API value
Cursor Ultra
Cursor · $200/mo
Tight
Over your $100 budget
Your need: $69–$238Included: 200–200 API-value $/mo included · community
Headroom bar: your need vs included capacity.
Smartest model
Opus 5 · intel 50.7
measured score · Cursor Agent + Tab
Windows
No 5h window; monthly pools (Cursor Models + Other Models)
Monthly included usage; priority access to new features
Tool integration
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's ownSubscription can't drive other agent harnesses; for other tools you'd pay per-token elsewhere.
$200/mo ≈ $0.667/task at 300 tasks/mo · API-value burn $0.23–$0.79/task across models
Up to $20/mo
Everyday coding. Claude Pro vs ChatGPT Plus vs Cursor Pro — or $20 of pure metered spend on OpenRouter / OpenCode Go credits.
Plan
Monthly
Smartest model
Included capacity
Tool integration
5-hour window
Caps & overage
ChatGPT Plus · ChatGPT
price: measuredcapacity: official
$20
5 6 Solintel 47.1 (measured)
10–100 Sol local msgs/5hofficial
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
10-100 Sol msgs/5h; Terra 25-200, Luna 250-2000 (official)
Weekly limits may also apply (amounts unpublished)+ credits / on-demand past the allowance
Claude Pro · Claude
price: measuredcapacity: official
$20
Opus 5intel 50.7 (measured)
40–80 Sonnet-equiv hrs/wkofficial
Harness lockedClaude Code, Claude Cowork, Claude Agent SDK (official Anthropic surfaces only)
At least 5x Free usage per 5-hour session
40-80 Sonnet-equiv hrs/wk (official); separate Opus weekly cap+ credits / on-demand past the allowance
Cursor Pro · Cursor
price: measuredcapacity: community
$20
Opus 5intel 50.7 (measured)
20–20 API-value $/mo includedcommunity
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's own
No 5h window; monthly pools (Cursor Models + Other Models)
Monthly included usage; on-demand top-up at API rates+ credits / on-demand past the allowance
OpenCode Go $20 · OpenCode Go
price: estimatedcapacity: derived
$20
4 6intel 44.4 (measured)
10–10 Zen credits $/mo usablederived
Works in any toolAny tool via OpenCode Zen OpenAI-compatible endpoints (chat/completions, responses, messages)
No windows; credit-metered
Credit-capped only; open coding models via Zen
OpenRouter $20 budget · OpenRouter
price: estimatedcapacity: derived
$20
Fable 5 1intel 53.4 (measured)
20–20 API-value $/mo budgetderived
Works in any toolAny OpenAI-compatible tool via API key (agents, CLIs, IDEs, scripts)
No windows; pure metered
Spend-capped only; pick any model per task
ChatGPT Plus: OpenAI positions Plus as 'a few focused coding sessions each week'. Credits can extend usage past the allowance. No annual billing.
Claude Pro: Includes Claude Code, Cowork, Design, Science. All usage shares one pool across web, desktop, mobile, and Code. Bursty users hit the 5h window before the weekly cap (community reports).
Cursor Pro: Unlimited tab completions, Bugbot, Cloud Agents. Best when you live in Grok/Composer; Opus-heavy use burns the $20 pool fast.
OpenCode Go $20: $20 total = $10 membership + $10 credits. Verdict compares usable credits against burn at Zen rates.
OpenRouter $20 budget: Budget comparison tier: what $20/mo buys at market per-token rates. Burn range spans Luna-class to Opus-class models.
Up to $50/mo
Small teams, or bigger metered budgets. ChatGPT Business or Cursor Teams Standard per seat — or $50 of tokens with no windows at all.
Plan
Monthly
Smartest model
Included capacity
Tool integration
5-hour window
Caps & overage
ChatGPT Business · ChatGPT
price: measuredcapacity: official
$25
5 6 Solintel 47.1 (measured)
10–100 Sol local msgs/5h (Standard)official
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
Plus-level Codex allowance on Standard; higher tiers ~Pro 5x (official)
Admin controls; workspace credits can extend access+ credits / on-demand past the allowance
Cursor Teams Standard · Cursor
price: measuredcapacity: community
$40
Opus 5intel 50.7 (measured)
40–40 API-value $/mo/seat includedcommunity
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's own
Per-seat monthly pool; centralized billing and admin
Team marketplace, Bugbot reviews, usage analytics, SSO+ credits / on-demand past the allowance
OpenCode Go $50 · OpenCode Go
price: estimatedcapacity: derived
$50
4 6intel 44.4 (measured)
40–40 Zen credits $/mo usablederived
Works in any toolAny tool via OpenCode Zen OpenAI-compatible endpoints (chat/completions, responses, messages)
No windows; credit-metered
Credit-capped only; open coding models via Zen
OpenRouter $50 budget · OpenRouter
price: estimatedcapacity: derived
$50
Fable 5 1intel 53.4 (measured)
50–50 API-value $/mo budgetderived
Works in any toolAny OpenAI-compatible tool via API key (agents, CLIs, IDEs, scripts)
No windows; pure metered
Spend-capped only; pick any model per task
ChatGPT Business: Per seat: $20/mo billed annually ($25 monthly). Minimum 2 seats. SSO, admin controls, no training on business data.
Cursor Teams Standard: Per seat per month. Cursor Router picks models per Auto request on Teams.
OpenCode Go $50: $50 total = $10 membership + $40 credits. Verdict compares usable credits against burn at Zen rates.
OpenRouter $50 budget: Budget comparison tier: what $50/mo buys at market per-token rates. Burn range spans Luna-class to Opus-class models.
Up to $100/mo
All-day agent coding for one person. Claude Max 5x, ChatGPT Pro 100, or Cursor Pro Plus — or $100 of metered spend.
Plan
Monthly
Smartest model
Included capacity
Tool integration
5-hour window
Caps & overage
Cursor Pro Plus · Cursor
price: measuredcapacity: community
$60
Opus 5intel 50.7 (measured)
60–60 API-value $/mo includedcommunity
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's own
No 5h window; monthly pools (Cursor Models + Other Models)
Monthly included usage; on-demand top-up at API rates+ credits / on-demand past the allowance
ChatGPT Pro 100 · ChatGPT
price: measuredcapacity: official
$100
6 Astraintel 52.8 (measured)
50–500 Sol local msgs/5hofficial
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
50-500 Sol msgs/5h (5x Plus, official)
Weekly limits may also apply (amounts unpublished)+ credits / on-demand past the allowance
Claude Max 5x · Claude
price: measuredcapacity: official
$100
Fable 5 1intel 53.4 (measured)
140–280 Sonnet-equiv hrs/wkofficial
Harness lockedClaude Code, Claude Cowork, Claude Agent SDK (official Anthropic surfaces only)
5x Pro usage per 5-hour session
140-280 Sonnet + 15-35 Opus hrs/wk (official); priority access+ credits / on-demand past the allowance
OpenCode Go $100 · OpenCode Go
price: estimatedcapacity: derived
$100
4 6intel 44.4 (measured)
90–90 Zen credits $/mo usablederived
Works in any toolAny tool via OpenCode Zen OpenAI-compatible endpoints (chat/completions, responses, messages)
No windows; credit-metered
Credit-capped only; open coding models via Zen
OpenRouter $100 budget · OpenRouter
price: estimatedcapacity: derived
$100
Fable 5 1intel 53.4 (measured)
100–100 API-value $/mo budgetderived
Works in any toolAny OpenAI-compatible tool via API key (agents, CLIs, IDEs, scripts)
No windows; pure metered
Spend-capped only; pick any model per task
Cursor Pro Plus: The natural tier for daily agent coding on mixed models.
ChatGPT Pro 100: New $100 tier (official help center): 5x Plus usage with full Astra access. No annual billing.
Claude Max 5x: Billed monthly. For all-day Claude Code sessions. Fable burns ~2x Opus (community N=1 reports: a heavy Fable day can eat a week's Fable credits).
OpenCode Go $100: $100 total = $10 membership + $90 credits. Verdict compares usable credits against burn at Zen rates.
OpenRouter $100 budget: Budget comparison tier: what $100/mo buys at market per-token rates. Burn range spans Luna-class to Opus-class models.
Over $100/mo
Heavy agent workloads. Claude Max 20x vs ChatGPT Pro 20x vs Cursor Ultra at $200, plus Cursor Teams Premium per seat.
Plan
Monthly
Smartest model
Included capacity
Tool integration
5-hour window
Caps & overage
Cursor Teams Premium · Cursor
price: measuredcapacity: community
$120
Opus 5intel 50.7 (measured)
120–120 API-value $/mo/seat includedcommunity
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's own
Per-seat monthly pool; centralized billing and admin
Team marketplace, Bugbot reviews, usage analytics, SSO+ credits / on-demand past the allowance
ChatGPT Pro 20x · ChatGPT
price: measuredcapacity: official
$200
6 Astraintel 52.8 (measured)
200–2,000 Sol local msgs/5hofficial
Works in any toolCodex CLI, Codex IDE extension, ChatGPT desktop; 'sign in with ChatGPT' works in many third-party agent tools
200-2000 Sol msgs/5h (20x Plus, official)
Weekly limits may also apply (amounts unpublished)+ credits / on-demand past the allowance
Claude Max 20x · Claude
price: measuredcapacity: derived
$200
Fable 5 1intel 53.4 (measured)
560–1,120 Sonnet-equiv hrs/wkderived
Harness lockedClaude Code, Claude Cowork, Claude Agent SDK (official Anthropic surfaces only)
20x Pro usage per 5-hour session
~4x Max 5x by tier ratio (derived); priority access+ credits / on-demand past the allowance
Cursor Ultra · Cursor
price: measuredcapacity: community
$200
Opus 5intel 50.7 (measured)
200–200 API-value $/mo includedcommunity
Vendor tools onlyCursor IDE, Cursor CLI, Cursor Cloud Agents — the harness is Cursor's own
No 5h window; monthly pools (Cursor Models + Other Models)
Monthly included usage; priority access to new features+ credits / on-demand past the allowance
Cursor Teams Premium: Per seat per month for heavy team agent use.
ChatGPT Pro 20x: Top ChatGPT tier plus Spark research preview. When capped, Chat falls back to weaker models automatically. No annual billing.
Claude Max 20x: Billed monthly. Top individual tier for heavy agent coding and parallel instances.
Cursor Ultra: For agent power users and multi-agent/automation workloads.
How the advisor works
Each provider caps usage in its own unit, so plans are judged natively: Claude in Sonnet-equivalent hours per week (Pro 40–80, Max 5x 140–280 + 15–35 Opus, Max 20x derived 4× — all from Anthropic's own announcements), Cursor in API-dollar pools per month (included value ≈ fee, community consensus; mechanism official), ChatGPT in Codex messages per 5-hour session (Plus Sol 10–100 through Pro 20x 200–2000, official Codex pricing).
Your need is converted per task size with stated assumptions: Q&A = 1 msg / 0.1h, coding task = 3 msgs / 0.5h, agent loop = 8 msgs / 1.5h, deep work = 12 msgs / 2.5h; steady rhythm spreads load over ~20 sessions/mo, bursty over ~8. Verdicts: need within the low end = Fits, inside the range = Tight, above it = Over.
Metered budgets are the no-subscription baseline: OpenRouter $20/$50/$100 buys that much API value at market rates with no windows, and OpenCode Go $20/$50/$100 means a $10 membership plus the rest as Zen credits priced at published Zen rates (burn uses the Zen offer tariffs, not merged prices). They usually win for light or spiky use and lose once a flat plan's included pool covers the month.
Smartest-model intelligence comes from measured Artificial Analysis indexes in the matrix; Cursor burn ranges come from measured per-token tariffs. When a plan caps you, Claude switches models down, ChatGPT falls back to weaker models, and Cursor continues on on-demand billing — a Tight verdict with a fallback is often fine for mixed work.
Tool integration is the third big factor after cost and intelligence. Open plans (ChatGPT via "sign in with ChatGPT", OpenRouter, OpenCode Go) drive any OpenAI-compatible harness; vendor-tools-only plans (Cursor) work only inside the vendor's own apps; harness-locked plans (Claude) officially support only Claude Code, Cowork, and the Agent SDK — using subscription auth in third-party harnesses violates Anthropic's ToS and risks account bans. Flip "Must work in third-party tools" to drop locked plans from the recommendations, and check the Most-portable-fit card.
Ranking priority: monthly cost (budget filter first), then intelligence (Effiq Score of each plan's smartest model on the coding profile), then portability — same-fee ties go to the more portable plan.
Evidence labels: measured (vendor list prices, benchmark scores, tariffs), official (vendor-stated caps), community (user-reported ranges: Cursor included value, Fable burn), derived (tier ratios, budget-tier math). No API key on your end would add firmer data — vendors publish no quota API.