Tony Wang5 min readClaude Code vs Codex vs Gemini CLI: Usage Limits Compared
Claude Code, Codex, and Gemini CLI meter usage three different ways in 2026 — here's how their actual limits and pricing compare, side by side.
Anthropic's Claude Code, OpenAI's Codex, and Google's Gemini CLI all meter usage differently, and all three have changed their limits at least once in 2026. If you're choosing between them — or budgeting for one already in use — the actual mechanics matter more than any single headline number, because the headline number keeps moving. Here's how the three compare as of August 2026, and what's actually changed under each one this year.
Claude Code: two overlapping windows, four changes since December
Claude Code is free to start, then Pro ($17–20/month), Max 5x and Max 20x (from $100/month), Team ($20–25/seat/month, Premium seats $100/seat/month annual), and Enterprise (custom). Usage is metered on two overlapping windows — a rolling 5-hour window and a 7-day window — shared across Claude Code, Claude.ai chat, and Cowork, so heavy use in one surface eats into the others.
The limit itself hasn't been stable:
| Date | Change |
|---|---|
| Dec 25–31, 2025 | Limits doubled as a holiday promotion |
| Jan 1, 2026 | Promotion expired; developers mistook the reversion for a cut |
| Mar 2026 | Peak-hour (5–11am PT) limit reductions introduced |
| Mar 31, 2026 | Anthropic publicly admitted users were hitting limits "way faster than expected" |
| May 6, 2026 | 5-hour limits doubled, peak-hour reduction removed, tied to a new SpaceX compute deal |
The weekly cap was not lifted in the May increase, and the Free plan got nothing from it — the improvements went to Pro and above.
Codex: bundled into ChatGPT, limits toggled on demand
Codex isn't sold on its own — it rides on ChatGPT's existing tiers: Free, Go ($8/month), Plus ($20/month), Pro 5x/20x ($100–200/month), Business ($20/user/month, 2-user minimum), Enterprise/Edu, plus pay-as-you-go API access. Historically it ran the same two-layer pattern as Claude Code — a rolling 5-hour window plus a separate weekly cap, shared between the local CLI and cloud Codex sessions.
On July 12, 2026, OpenAI temporarily removed the 5-hour window entirely for Plus, Pro, and Business plans, leaving only the weekly cap — a response to a traffic spike after a model launch a few days earlier, explicitly labeled temporary with no announced end date. It's the same week Codex hit a reported 6M-then-7M active users — a figure OpenAI's own posts described as Codex and ChatGPT Work combined, not Codex alone.
Gemini CLI: the only genuinely free daily quota
Gemini CLI is free via a Google account login (Gemini Code Assist for Individuals) — roughly 60 requests/minute and 1,000 requests/day, no card required, with a 1M-token context window. A Gemini API key carries the same 1,000/day free ceiling. Paid Code Assist tiers exist for teams needing higher quotas, SLAs, or admin controls (Standard around $22.80/user/month annual, Enterprise around $45/user/month), and Vertex AI is the enterprise billing path with the highest limits available.
Side by side
| Claude Code | Codex | Gemini CLI | |
|---|---|---|---|
| Free tier | Yes, limited | Yes (ChatGPT Free) | Yes — 1,000 req/day, no card |
| Entry paid tier | Pro, $17–20/mo | Plus, $20/mo | Standard, ~$22.80/user/mo |
| Limit structure | Rolling 5-hour + 7-day window | Rolling 5-hour window + weekly cap (currently: weekly only) | Requests/minute + requests/day |
| Changed in 2026? | 4 times (Dec, Jan, Mar, May) | Yes (5-hour window removed Jul 12) | No major public change found |
| Most transparent about limits | Exposes a rate-limits API with live percentages | Weekly reset dates communicated via social posts | Fixed, documented per-minute/per-day caps |
What this actually means for developers
If you want to try one without spending anything first, Gemini CLI's free daily quota is the lowest-friction entry point — no card, no plan to pick, no ambiguity about what counts against your limit.
If you're already paying for one and hitting limits, the pattern in 2026 has been that both Anthropic and OpenAI raise limits reactively, in response to visible user complaints or demand spikes, not on a published schedule. That's good news if you're currently frustrated — a fix is plausible — but it also means today's generous limit (Codex's suspended 5-hour window, in particular) is explicitly provisional and could revert without warning.
If you're budgeting for a team, Claude Code's shared window across Claude Code, chat, and Cowork is worth accounting for explicitly — a team using all three surfaces will hit the ceiling faster than usage of any single surface alone would suggest.
None of this is a verdict on which tool writes better code — that's a separate question from what these three actually let you run before you hit a wall.
Sources
Related reading
- Did Codex Really Overtake Claude Code? The 6M (Now 7M) User Number, Fact-Checked — what that widely-shared user-count comparison actually measured.
- Claude Code vs Codex: Usage Limits Compared (July 2026) — the specific July 12–13 limit changes this post builds on.
- Web scraping with AI — using these same coding agents for data-extraction workflows.
Every figure in this post came from each vendor's own pricing or announcement pages, pulled directly rather than secondhand — if you want to track how these limits shift, the X profile/post and Google Trends endpoints in the Playground are the same tools used to verify the sibling post on this topic.
Frequently asked questions
Which AI coding tool has the most generous free tier?
Gemini CLI — it's free via a Google account with roughly 1,000 requests/day and no card required, the most generous no-signup-friction entry point of the three. Claude Code and Codex both offer free tiers too, but with tighter, less transparent limits.
Why do Claude Code's usage limits keep changing?
Anthropic adjusted Claude Code's limits four times between December 2025 and May 2026 — a holiday promotion, a peak-hour throttle, a public admission that users were hitting limits faster than expected, and a doubling tied to a new SpaceX compute deal. The pattern has been reactive, not scheduled.
Did OpenAI remove Codex's usage limit?
Partially, and temporarily. On July 12, 2026, OpenAI removed Codex's rolling 5-hour window for Plus, Pro, and Business plans, leaving only the weekly cap — a response to a demand spike, explicitly labeled temporary with no announced end date.
Is Gemini CLI good enough for daily coding work?
For solo, moderate use, its free 1,000 requests/day and 60-requests/minute caps are generous enough for regular use. Teams needing higher throughput, SLAs, or admin controls should look at the paid Code Assist tiers or Vertex AI instead.
Which tool is best for a team, not a solo developer?
Claude Code's shared limit across Claude Code, Claude.ai chat, and Cowork is worth budgeting for explicitly on a team that uses all three surfaces. Codex's Business plan and Gemini's Enterprise Code Assist both offer team-oriented pricing with higher, more predictable quotas than their free tiers.