What Gemini’s New Token Limits Change for Users
Gemini’s new token limits are compute-based quotas that measure the complexity, length, and features used in each interaction instead of counting a fixed number of prompts per day, reshaping how users experience API usage quota and AI rate limiting across Google’s tiers. Under the updated system, Gemini plans—including AI Plus, AI Pro, and AI Ultra—share access to models like 3.5 Flash and 3.1 Pro, but heavy users report hitting the ceiling much faster. Google moved away from daily prompt caps to a rolling five-hour limit that refreshes until a broader weekly quota is reached. This aligns Gemini token limits more closely with rivals such as Claude, yet users say it feels harsher, especially when 3.5 Flash seems less reliable than 3.1 Pro. For power users who rely on long conversations or complex prompts, the new model turns every session into a calculation about cost, compute, and risk.
Unexpected Quota Caps and a Single-Prompt Meltdown
The most visible friction with Gemini token limits comes from users hitting caps far sooner than expected. One Google AI Pro subscriber reported that a single failed video-generation request using the avatar feature consumed their entire five-hour allowance in a matter of minutes, highlighting how fragile the new system can feel. According to Android Authority, the user started at 0% usage, ran the prompt for three to four minutes, then watched it fail while the rate limit jumped to 100%. Google’s Josh Woodward responded publicly, saying, “Yikes, let us take a look!”, which at least signals that the issue is on the company’s radar. But the episode reinforces a broader complaint: when one misfire can burn a five-hour window, the balance between infrastructure protection and user experience looks off, especially for paying customers who expect stable Gemini Pro pricing and predictable usage.
Antigravity Quota Boosts and Persistent Frustration
As backlash grew, Google moved to ease pressure on Gemini token limits—at least inside its Antigravity interface. After Reddit threads accused the company of a bait-and-switch on the AI Pro plan, Google DeepMind director Varun Mohan announced that Antigravity users would see their Gemini rate limits tripled and weekly quotas reset. When criticism continued, he followed with another update, tripling those weekly quotas again. According to Android Authority, this works out to roughly a 9x increase from the post-nerf state, but only within Antigravity; broader Gemini caps remain unchanged. Many power users still say limits are tighter than before Google’s quiet rollback. The episode underlines how opaque AI rate limiting policies can erode trust: developers feel they are testing how far they can push their workflows before the quota hammer falls, instead of focusing on building and shipping.
From Open-Source Gemini CLI to Closed Antigravity
Alongside token policy changes, Google is pushing Pro, Ultra, and free users away from the open-source Gemini CLI toward the closed-source Antigravity CLI, a move that adds another layer of tension. The older Gemini CLI, hosted on a busy GitHub repository, enabled custom workflows and community contributions. By contrast, Antigravity’s repository is sparse, and Google has confirmed there will not be “1:1 feature parity right out of the gate.” Developers now face fewer features, lack of open-source transparency, and stricter usage controls in the same transition. One Redditor summed up the anxiety: they had “all kinds of custom layers” on top of Gemini CLI and are “quite anxious about usage limits.” Image: An illustration of developers adapting to Google’s Antigravity CLI shift. For teams that depend on flexible tooling plus generous API usage quota, the migration feels like losing both control and capacity at once.
How Gemini Compares to Competitors and What’s at Stake
Gemini’s new token limits invite direct comparison to rivals like Claude, which also use aggressive AI rate limiting to protect infrastructure costs. Critics argue that Google’s compute-based quotas might be “just as bad as Claude’s, and maybe even a little dumber” because they restrict access to older but more reliable models such as 3.1 Pro while promoting newer options like 3.5 Flash that some testers find less consistent. Subscription reshuffles add to the confusion: AI Plus starts at USD 7.99 (approx. RM37), AI Pro remains at USD 19.99 (approx. RM93) per month, and AI Ultra’s highest tier dropped from USD 250 (approx. RM1,164) to USD 200 (approx. RM931), with a new USD 100 (approx. RM466) option. The question is whether these Gemini Pro pricing tweaks offset the friction of hitting token caps mid-workflow. For now, developers see limits rising in Antigravity but not across the stack, and wonder if cost control is winning out over trust and reliability.