Pricing landscape · 2026-08-22

AI Coding Tool Pricing 2026
Four at $20, three at $100

Cursor, Claude Code, Antigravity and Windsurf all sit at $20/mo; the heavy tiers line up at $100. With prices converged, the real question is subscription vs pay-as-you-go.

#cursor pricing#claude code pricing#antigravity pricing#pay-per-use vs subscription

Price map (2026-08 snapshot)

$20

Converged entry tier

Cursor Pro, Claude Pro (including Claude Code) and Windsurf Pro are all $20/mo; Google AI Pro is $19.99.

$100

Converged heavy tier

Claude Max 5x, Google AI Ultra and ChatGPT Pro are all $100/mo. Heavy users are effectively pushed to this tier.

$0.20

Lowest pay-per-use price (GPT-5.6 Luna input)

After the 07-30 price cut, Luna input is $0.20 per million tokens. Light tasks on pay-per-use often cost under $1 a day.

$9.99

BYOK hosted entry point

Hosted inference like ClinePass keeps lowering the BYOK barrier; the extension itself is free.

How the convergence happened

In 2026, AI coding tool pricing completed a round of convergence: entry tiers all aligned at $20/mo (Windsurf only caught up from $15 in May), and heavy tiers all aligned at $100/mo. Price is no longer the differentiator — quota policies, model lists and agent experience are. For users, the question changed from 'which is cheapest' to 'does a subscription or pay-per-use fit me better.'

This month's new variables

On 07-30, OpenAI slashed GPT-5.6 Luna by 80% (input at $0.20 per million tokens) and Terra by 20%, sharply raising the value of the pay-per-use route; OpenRouter and others quickly followed with extra discounts. On the other side, Anthropic briefly planned (in June) to move Claude Code's programmatic calls to API billing — paused, but the direction is clear: the boundary between subscription and metered is blurring.

Timeline

2026-05

Windsurf Pro rises from $15 to $20, completing the entry-tier alignment.

2026-07-30

OpenAI cuts prices: Luna -80%, Terra -20%, Sol unchanged; Sol gains a 2.5x Fast mode (at 2x the price).

2026-08

The two-layer structure — four at $20, three at $100 — stabilizes; pay-per-use vs subscription becomes the main selection axis.

Confirmed vs caveats

Confirmed

All prices above come from official pricing pages or official announcements (fetched 2026-08). The Luna/Terra cuts are corroborated by OpenAI updates and multiple media reports.

Caveats

'$20 for unlimited use' is a misconception: every $20 tier has rate and quota limits, and heavy agent usage easily hits the ceiling — which is exactly why the $100 tier exists. When comparing, look at the quota terms, not just the monthly fee.

Subscription vs pay-per-use

Subscription ($20/$100 tiers)

Predictable budget, zero hassle. Suits people who use a single tool heavily and steadily every day. Usage beyond quota is billed extra or throttled.

Pay-per-use (API billing)

Pay for what you burn, switch models freely. Suits people with fluctuating usage, multi-tool setups, or a need to mix several model families.

How to run your own numbers

First estimate monthly tokens: light use (a few chat-level calls a day) often burns under $5 a month — pay-per-use wins outright; moderate use (a few hours of pair programming daily) lands near break-even with the $20 tier; only heavy use (all-day multi-agent) justifies $100. Most people's mistake is assuming they're heavy users.

On QCode

QCode is the base layer of the pay-per-use route: one key reaches seven families, at official pricing times our service rate, with no subscription and no minimum. Pair it with Cline / OpenCode / Claude Code and route chores to Luna-class cheap models while flagships handle the hard tasks — the cost stays under your control.

FAQ

What do AI coding tools cost in 2026?

Entry tiers: Cursor Pro $20, Claude Pro $20, Google AI Pro $19.99, Windsurf Pro $20. Heavy tiers: Claude Max 5x $100, Google AI Ultra $100, ChatGPT Pro $100.

Is the $20 tier enough?

Depends on usage. Conversational coding is fine; all-day parallel agents are not — every tier has rate and quota terms, and heavy usage gets pushed toward the $100 tier.

What does pay-per-use cost per month?

Light users often stay under $5; moderate users land around $20; only all-day heavy agents exceed $100. The safest approach: run pay-per-use for a week, then decide whether to subscribe.

Is pay-per-use a better deal after the GPT-5.6 price cut?

Yes. Luna input is $0.20 per million tokens (-80%), Terra is $2/$12 (-20%). Route chores to Luna-class models and the whole bill can drop another order of magnitude.

Can subscription and pay-per-use be mixed?

Yes — and it's common: the subscription tool covers the main IDE experience, while pay-per-use APIs cover automation scripts, CI and the second tool. They don't conflict.

How does QCode bill?

No subscription: official pricing times our service rate, per token, with usage and spend visible live in the console.

Sources

Official pricing pages of each vendor (fetched 2026-08), OpenAI's 07-30 price-cut announcement plus CNBC coverage, FindSkill comparison posts, No Code MBA.

Run pay-per-use for a week before you subscribe

QCode has no minimums and no subscription — official pricing times our service rate, so real usage answers 'should I buy the $20 tier.'