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.
Price map (2026-08 snapshot)
Converged entry tier
Cursor Pro, Claude Pro (including Claude Code) and Windsurf Pro are all $20/mo; Google AI Pro is $19.99.
Converged heavy tier
Claude Max 5x, Google AI Ultra and ChatGPT Pro are all $100/mo. Heavy users are effectively pushed to this tier.
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.
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
Windsurf Pro rises from $15 to $20, completing the entry-tier alignment.
OpenAI cuts prices: Luna -80%, Terra -20%, Sol unchanged; Sol gains a 2.5x Fast mode (at 2x the price).
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.'
Related reading
QCode pricing explained
The service rate and billing terms.
Claude Code cost optimization
Practical ways to drop the bill by an order of magnitude.
AI coding tools comparison
The dimensions beyond price.
Prices are a 2026-08 snapshot; the official pricing pages prevail. Not affiliated with any vendor listed.