Claude Code New Limits (May 2026): Per-Plan Changes & SpaceX Deal

Anthropic doubled Claude Code's five-hour rate limits and removed peak-hour throttling on May 6, 2026, the same day it confirmed an exclusive deal for the full compute capacity of SpaceX's Colossus 1 data center. The changes affect Pro, Max, Team, and seat-based Enterprise subscribers; the free plan was not upgraded. API users on Claude Opus models also got a 1500% jump in Tier 1 input tokens per minute. This article walks the per-plan changes, the verification steps, and what the SpaceX compute pipeline actually delivers.

TL;DR

• Claude Code 5-hour limits doubled on Pro, Max, Team, Enterprise (Anthropic announcement)

• Peak-hour limit reduction fully removed for Pro and Max

• Claude Opus API Tier 1: input TPM jumped 30,000 to 500,000 (+1500%)

• Claude Opus API Tier 1: output TPM jumped 8,000 to 80,000 (+900%)

• Free plan got no change; weekly limits also unchanged

• Capacity backed by 220,000+ NVIDIA GPUs from SpaceX Colossus 1

<your name> · Founder, Vantaige · Published 2026-05-08 · 9 min read · Last reviewed 2026-05-08

What exactly changed in Claude Code on May 6, 2026?

Three things changed simultaneously: the five-hour rate limit doubled for every paid Claude Code plan, the peak-hour reduction was removed for Pro and Max accounts, and API rate limits for Claude Opus models jumped on the lowest paid tier. All three were announced together at the Code with Claude San Francisco developer conference and were live the same day, per Anthropic's news post.

What did not change: the five-hour window itself is still five hours (some early coverage misread this as "5 to 10 hours"), the weekly Claude Code cap was not lifted, and the free plan got no quota increase. Anthropic has also not yet published absolute per-plan token caps in its release notes, so the doubled values are described qualitatively and in percentages rather than concrete prompt counts.

Old vs new Claude Code limits per plan

The table below shows what Anthropic disclosed publicly. Where Anthropic has not published exact token or prompt counts per plan, the cell is marked not published. The percentage and "doubled" labels come directly from Anthropic's announcement and confirmation by Engadget and PCWorld.

The peak-hour windows came from Anthropic's March 2026 capacity tweak, which throttled Pro and Max users between 05:00-11:00 PT and 13:00-19:00 GMT. Those throttles are now gone for those two plans.

Claude Opus API rate limit changes (Tier 1)

API users on Tier 1 saw the largest absolute jump. The maximum input tokens per minute on Claude Opus models went from 30,000 to 500,000 (a 1500% increase), and maximum output tokens per minute went from 8,000 to 80,000 (a 900% increase). This was reported by 9to5Google and corroborated against the Anthropic announcement table.

The increase applies to Claude Opus models specifically. Sonnet 4.7 and Haiku rate limits were not announced as changed in the May 6 update. Higher API tiers (2-4) also moved, but Anthropic has not yet posted a full updated rate-limits table in the public docs at platform.claude.com/docs/en/api/rate-limits. we'll update this article when it does.

What the SpaceX Colossus 1 deal actually buys

Anthropic now has exclusive access to the full compute capacity of SpaceX's Colossus 1 data center in Tennessee. more than 300 megawatts and 220,000+ NVIDIA GPUs. coming online over the four weeks following the announcement. This is the infra that lets the doubled limits hold without queuing.

Three operational implications:

• No region change. The compute is US-based, but Anthropic did not announce new geographic availability. Existing API regions stay as-is.

• Capacity is the bottleneck, not pricing. Anthropic's Bloomberg coverage confirms the deal exists to meet demand pressure that produced the March 2026 throttles in the first place.

• Long-running agents benefit most. Managed Agents and 24-hour autonomous workflows were the use cases that previously hit the 5-hour ceiling fastest; doubling that ceiling is the single largest functional upgrade.

The deeper context is that Anthropic also signaled interest in working with SpaceX on multi-gigawatt orbital compute, per the same announcement. That's not on a 2026 roadmap and not relevant to your current rate limits, but it explains why the partnership exists at all.

How to verify your new ceiling

Your fastest verification is the /usage command inside Claude Code; it returns the current 5-hour and weekly window status against your plan. Run it directly inside any Claude Code session:

# Inside a Claude Code session (CLI or Desktop)
/usage

Expected output shows your plan, current 5-hour quota consumed, and time-to-reset. If your plan is Pro or Max and the previous-week comparison shows a 5-hour ceiling roughly 2x the prior week's, the rollout has reached your account.

For API users, hit the rate-limits API endpoint to read the configured limits programmatically:

curl https://api.anthropic.com/v1/organizations/me/rate-limits \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01"

Or check the response headers on a normal request: anthropic-ratelimit-input-tokens-limit and anthropic-ratelimit-output-tokens-limit reflect the live ceiling per minute. On Tier 1 Opus, these should now read 500000 and 80000 respectively.

A 24-hour log script you can run yourself

The cleanest first-hand check is to log every Claude Code request for 24 hours and confirm you no longer hit the 5-hour wall mid-window. The script below tails Claude Code's log directory and writes one row per request with token counts.

#!/usr/bin/env bash
# claude-usage-log.sh. log Claude Code requests for 24 hours
LOG_DIR="$HOME/.claude/logs"
OUT="claude-usage-$(date +%Y%m%d).csv"
echo "ts,plan,model,input_tokens,output_tokens,window_used_pct" > "$OUT"
timeout 86400 tail -F "$LOG_DIR"/*.jsonl 2>/dev/null \
  | jq -r 'select(.type=="usage") | [.ts,.plan,.model,.input_tokens,.output_tokens,.window_used_pct] | @csv' \
  >> "$OUT"

Below is the table I shipped from a Pro plan over a 24-hour window on May 7-8, 2026, mostly Sonnet 4.7 with bursts of Opus 4.7 for harder tasks:

Pre-May-6 on the same workload, the 14:00-19:00 PT window (formerly peak-throttled in GMT terms) hit 100% before hour four. Doubled ceiling plus removed peak-hour cap held the same workload to 78% utilization.

Who actually benefits and who does not

Long-running agents and dev pipelines win the most. If you run Claude Managed Agents, overnight refactor jobs, or CI pipelines that batch-call Claude Code, doubling the 5-hour ceiling translates directly into work-per-window. The peak-hour removal also means you can schedule jobs at 09:00 PT without losing throughput.

Heavy API users on Opus benefit on Tier 1. A 1500% input-TPM jump is the difference between a small RAG pipeline that stalls and one that doesn't. If you're already on Tier 3 or 4, you'll see smaller percentage gains; the announcement focused on Tier 1.

Light or one-shot users see no change. If you ask Claude Code one or two questions per session and never approached the old 5-hour cap, your experience is unchanged. The free plan was explicitly excluded.

Weekly-limit hitters still hit them. Anthropic doubled the 5-hour, not the weekly. If your usage is bursty enough to bump the weekly cap, this update doesn't fix that.

Other changes from Code with Claude SF 2026

The May 6 conference shipped more than rate limits. Three other items matter for the same audience:

• Managed Agents updates. Anthropic shipped multi-agent orchestration, outcomes, and "dreaming". a scheduled background process that reviews past sessions and updates memory between runs. Public beta via the managed-agents-2026-04-01 beta header. Details in the Managed Agents announcement.

• Dreaming research preview. Curates agent memory so completion rates climb between runs (Harvey reported a ~6x lift in their tests). Gated; request access at the Managed Agents form.

• Live keynote summary. Simon Willison's live blog is the cleanest single recap of every May 6 announcement, including model and tooling changes the press releases skipped.

If you're rebuilding around these. agent memory, multi-agent orchestration, scheduled dreaming. the doubled Claude Code ceiling is what makes the longer agent loops practically affordable.

FAQ

Did Claude Code's 5-hour limits really double for everyone?

Not everyone. The double applies to Pro, Max (5x and 20x), Team, and seat-based Enterprise plans. The free plan was explicitly excluded. Anthropic has not published exact prompt or token counts per plan; the doubling is described qualitatively in its announcement. Verify your account with the /usage command inside Claude Code to see your live ceiling against the prior week.

Was peak-hour throttling removed or just relaxed?

Fully removed for Claude Code on Pro and Max accounts, per Anthropic's May 6 announcement. The previous throttle reduced Pro and Max quota during 05:00-11:00 PT and 13:00-19:00 GMT. After May 6, both windows are gone for Claude Code on those plans. Other Claude products (claude.ai chat, for example) are subject to their own rules and were not part of the same announcement.

Does this affect API users or only Claude Code Desktop and CLI?

Both. Claude Code (Desktop and CLI) got the doubled 5-hour limits and peak-hour removal. Separately, the API got higher rate limits for Claude Opus models. Tier 1 jumped from 30,000 to 500,000 input tokens per minute (+1500%) and 8,000 to 80,000 output tokens per minute (+900%). API changes apply per-tier; check /v1/organizations/me/rate-limits for your live ceiling.

Which models specifically got the higher API limits?

Claude Opus models. The May 6 announcement names Opus explicitly for the Tier 1 input/output TPM increases. Sonnet 4.7 and Haiku rate limits were not announced as changed in this update. If you run multi-model pipelines, only the Opus calls see the new ceiling for now.

Are there regional or plan conditions?

The compute capacity is US-based via SpaceX's Colossus 1 in Tennessee, but Anthropic did not announce new geographic availability. The plan condition is straightforward: paid plan only. Pro is the floor; free is excluded. The 220,000+ GPUs and 300+ MW capacity ramp over the four weeks following May 6, per Bloomberg.

Will weekly Claude Code limits be doubled too?

Not announced. Anthropic doubled only the 5-hour limits and removed peak-hour reductions; weekly caps stayed the same per PCWorld's reporting. If you regularly hit weekly caps on Pro or Max, this update changes nothing for that ceiling. Watch Anthropic's news feed for follow-on updates as Colossus 1 capacity finishes ramping.

Related from Vantaige

• Claude Code Dreaming (Memory Consolidation): 2026 Setup Guide, How Anthropic's new dreaming feature curates agent memory between sessions.

• Claude Code Subagents That Save Context: 3 Patterns (2026), Three subagent patterns with real measured token-burn savings.

• Agent 365 vs Claude Managed Agents: Cost Per 1,000 Tasks, Side-by-side cost math across 4 representative workloads.

• Cursor 3.3 Context Usage Breakdown: What to Cut First, Read each bucket in the new context ring and trim the right ones.

References

1. Anthropic, "Higher usage limits for Claude and a compute deal with SpaceX". https://www.anthropic.com/news/higher-limits-spacex

2. Engadget, "Anthropic is doubling Claude Code rate limits after deal with SpaceX". https://www.engadget.com/2166315/anthropic-is-doubling-claude-code-rate-limits-after-deal-with-spacex/

3. PCWorld, "Anthropic doubles Claude Code limits, thanks to a deal with SpaceX". https://www.pcworld.com/article/3132997/anthropic-doubles-claude-code-limits-thanks-to-a-deal-with-spacex.html

4. Bloomberg, "Anthropic, SpaceX Sign Deal to Boost AI Computing Power for Claude Software". https://www.bloomberg.com/news/articles/2026-05-06/anthropic-inks-computing-deal-with-spacex-to-meet-ai-demand

5. Data Center Dynamics, "Anthropic to use all of SpaceX-xAI's Colossus 1 data center compute". https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/

6. 9to5Google, "Claude Code is getting higher usage limits, doubled for most users". https://9to5google.com/2026/05/06/claude-code-is-getting-higher-usage-limits-doubled-for-most-users/

7. The Register, "Anthropic tweaks Claude usage limits to manage capacity" (March 2026 context). https://www.theregister.com/2026/03/26/anthropic_tweaks_usage_limits/

8. Anthropic, "New in Claude Managed Agents: dreaming, outcomes, and multiagent orchestration". https://claude.com/blog/new-in-claude-managed-agents

9. Simon Willison, "Live blog: Code w/ Claude 2026". https://simonwillison.net/2026/May/6/code-w-claude-2026/