Google Antigravity Credits: What You Actually Pay After the March 2026 Protest

One developer hit their weekly quota limit after 20 minutes of work. The Register reported they’d gone from 300 million input tokens per week to 9 million (a 97% cut) and Google’s original docs promised “high, generous quotas.” No more.

Google launched Antigravity in November 2025 as the in-IDE agent competitor to Cursor and Windsurf. For four months the pitch held up: accessible, agentic coding environment, reasonable free tier. Then March 2026 arrived and the company introduced a credit-based billing system that reframed what “accessible” actually meant. Developers who’d built Antigravity into their daily workflow woke up locked out.

What Antigravity Actually Is

Antigravity is Google’s agentic IDE. Think Cursor’s agent mode built into a web-based editor. You point it at code, describe what you want changed, and the agent either refactors in-place or generates new modules. It integrates with Google’s Gemini 3.1 models and promises token-efficient context windowing for large codebases.

When Google shipped it in November 2025, the positioning was deliberate: an open-source-friendly, quota-generous alternative to Cursor Pro ($20/mo) and Windsurf (similar price, similar positioning). Antigravity’s free tier started with 250 agent requests per day. That detail matters later.

The Three Plans, Stated Simply

Google’s current Antigravity lineup is:

• Free: Zero cost, 20 agent requests per day (down from 250 in December), weekly quota reset
• AI Pro: $20/month, higher request limits, quota refresh every five hours (in theory; see below)
• AI Ultra: $249.99/month, highest token allocations, claimed priority queuing

The credit system runs parallel: $25 buys 2,500 credits, or roughly $0.01 per credit. Bulk purchase: $199 for 20,000 credits. When your included quota exhausts, credits burn next.

What a Credit Actually Costs in Practice

Here’s where the transparency broke down. Google’s docs never published a credit-to-token conversion table. One developer burned 2,500 credits (the base purchase unit) in under two hours of normal agent usage, suggesting credits are expensive relative to what you’d expect from the $20/mo price point.

We ran through the math assuming shorter, routine tasks (roughly 15-minute agentic sessions rather than full 2-hour refactors): five sessions per day across a five-day work week runs roughly $60 to $80 in overages above the plan cost. That’s not an edge case for anyone using Antigravity as a primary IDE; it’s the expected trajectory for routine coding work.

The March 2026 Quota Cut

On March 12, 2026, The Register broke the story. Developers on the Google AI forum had been posting for weeks: their AI Pro quotas were not refreshing every five hours anymore. They refreshed once weekly. And when they hit the weekly ceiling, the IDE locked them out for the remainder of the week, sometimes 167 hours of zero access.

One tracked case: previously 300+ million input tokens per week, 1 to 2 million output. The following week: under 9 million input, 200 thousand output. That is a 97% reduction in input capacity.

DevClass covered the same incident with identical reporting: the refresh window changed, quotas plummeted, and forum threads like “Workflow completely blocked” and “Unacceptable Antigravity quotas” flooded the forums.

Why Developers Are Screaming

The anger wasn’t really about the quotas themselves. It was about the invisibility.

Google’s original launch docs used vague language: “high quotas,” “generous limits.” The company never published what those meant in tokens, requests, or cost. Then when the change hit, there was no announcement. Developers woke up to locked accounts and support threads that Google didn’t respond to for days.

The second complaint: no credit-to-token documentation. Developers couldn’t calculate cost-per-agent-session before it happened. Ultra users (paying $250/mo) discovered they also hit hard quota caps, suggesting the tiers don’t work as advertised.

The third: the shift from every-five-hours refresh to once-weekly made the product unusable for iterative development. Five-hour resets meant you could test, hit a limit, wait, and continue the same day. Weekly resets meant work stoppages.

How It Stacks Up Against Cursor Pro and Windsurf

Cursor Pro is $20/month with unlimited requests and a 10-message-per-day quota on Claude 3.5 Sonnet (the actual bottleneck). You can send 10 full-context agentic prompts, then refresh the next day.

Windsurf is similar: subscription-based, predictable limits, no surprise credit burns.

Antigravity at $20/mo looks cheaper until you factor in overages. A developer doing 5 to 10 agent sessions per day, standard for someone using Antigravity as a primary IDE, will blow past the quota by mid-week and face either a $60 to $80 weekly overage or a $250 upgrade to Ultra (which also has caps, just higher ones).

For an apples-to-apples monthly cost at moderate usage (five agent sessions per day, 20 sessions per week): Cursor Pro comes in at $20/mo flat. Windsurf is similar. Antigravity? $20 plus $60 to $80 in weekly overages translates to roughly $80 to $100 per month effective cost, if you’re willing to eat the surprise charges. Or $250/mo if you want to avoid the guessing game entirely.

Our Take

Antigravity had a real shot at owning the “open-source-friendly, affordable agentic IDE” space. The product is solid. The pricing model undercut Cursor and Windsurf on paper.

But Google executed a classic bait-and-switch: launch cheap to build adoption, introduce opacity (vague docs, hidden credit economics), then use that opacity as cover for quota cuts that funnel power users toward the $250 plan. It’s not illegal. It’s just the move that kills developer trust.

When $20/mo is defensible: if you’re in the free tier or doing light editing (fewer than three agent sessions per week). The risk is low, the feature set is real, and you’re not paying for overages.

When it’s not: if you’re a daily power user. The effective cost balloons, the monthly bill becomes unpredictable, and you’re paying for the privilege of using a product you expected to be affordable. At that point, Cursor Pro or Windsurf at a flat $20/mo with transparent quotas is the rational choice.

We’ve tracked a dozen AI tool pricing pivots this year. Antigravity’s is one of the more cynical ones. Google’s selling this as a platform for developers. The pricing says otherwise.