Claude

Last verified: Jun 6, 2026

1 unannounced

20 sourced changes·90% official·verified against official docs, changelogs & Wayback·last checked Jun 6, 2026

01 / VERDICT

Watch

Some recent constraint changes — monitor before depending on it.

WHY WATCH — trend: tightening · drift -10% over 90d · 2 tightenings in 90d

Constraint Stability

80/100

higher = fewer recent restrictions

tightening-10% / 90d

1 unannounced change

Claude has become more restrictive over time, mainly in model availability.

Most noticeable: Anthropic's newest and most capable models became unavailable to every customer overnight by federal directive — users building on Fable 5 / Mythos 5 must migrate with no notice. First known instance of the US government forcing a deployed frontier model offline.

Recency

days since it last tightened

Velocity

2.8

tightenings per year

Silent

25%

of tightenings were unannounced

Reversals

75%

of tightenings later reversed (same dimension)

Claude constraint metrics — computed from 4 tracked tightenings · median 81 days between tightenings. Original to this site; not published by the vendor.

IMPACT BY PERSONA

Solo / occasional

Brief, occasional use is rarely touched by limit changes.

low

Power user / daily heavy

Daily-limit, session and rate constraints have been steady.

low

Team / shared workflow

Shared quotas, model availability and per-seat cost have held steady.

low

Startup / cost-scaling

Pricing and usage-based cost have been predictable as volume grows.

low

Enterprise / long-term dependency

Continuity exposure: ownership / policy shifts.

medium

02 / ANALYSIS

In August 2025 Anthropic layered two seven-day weekly caps on top of the existing rolling five-hour limits, explicitly to curb heavy Claude Code usage — the structural tightening that anchors this record. Since then the limits have swung both ways: a temporary holiday doubling over the 2025 turn of year, then a quieter and more concerning move in March 2026, when weekday morning allowances began draining faster. That silent, peak-hour throttle was only later confirmed. By May 2026 it had been reversed and five-hour limits doubled for Pro and Max. Transparency here is mixed — reset logic was eventually fixed to a predictable weekly schedule, but the one unannounced change is a meaningful mark against it. The sharpest event was not Anthropic's doing at all: in June 2026 a US government export-control directive forced Fable 5 and Mythos 5 offline for every customer worldwide overnight, the first known instance of a deployed frontier model being pulled by federal order, with access returning July 1. Model and price changes otherwise trend neutral-to-favorable, including Sonnet 5's introductory pricing below its predecessor. In short, Claude's limits are actively managed and have loosened net over time, but they are adjusted frequently, occasionally without announcement, and — uniquely here — a specific model's availability can vanish for reasons entirely outside the vendor's control. Build on the platform, but do not hard-wire a dependency on any single named model.

03 / CURRENT LIMITS

Current state + recent changes

Free: Sonnet-class, dynamic ~5-hour sliding window, no guaranteed message count
Pro ($20/mo): At least 5x Free per session; 5-hour reset + a weekly cap across models
Max ($100 / $200): 5x / 20x Pro usage; two weekly limits (all-models + Sonnet-only)

Last verified: Jun 6, 2026

RECENT LIMIT & PRICING CHANGES

BeforeAfter

N/A↑Pricing page now lists context window per tier: 500k on default model for Enterprise (self-serve and sales-assisted); 200k for Free, Pro, Max and Team.

Jul 2026·source

Flagship coding slot: Sonnet 4.6 at Input $3 / Output $15 per MTok→Sonnet 5 takes the flagship coding slot at introductory $2 / $10 per MTok (input/output) through Aug 31, 2026, then $3 / $15 standard; Sonnet 4.6 moves to a secondary listing (unchanged $3 / $15).

Jun 2026·source

Top API tier was Opus at $5 / 1M input, $25 / 1M output→New Fable 5 tier above Opus at $10 / 1M input, $50 / 1M output (Opus pricing unchanged)

Jun 2026·source

Peak-hour depletion + standard 5h caps↑5-hour limits doubled for Pro/Max; peak-hour reduction removed (+25%)

May 2026·source

04 / WHERE IT MATTERS

Which usage patterns the recent changes affect — each line backed by the events driving it.

Specific-model dependencehigh exposure

Aug 2026↓Anthropic announced tightened biology-domain safeguards for Fable 5; announcement headline only, no published detail on which requests are affected.·source

Jun 2026↑Restores access to Fable 5 model previously suspended due to export controls.·source

Jun 2026↓Anthropic's newest and most capable models became unavailable to every customer overnight by federal directive — users building on Fable 5 / Mythos 5 must migrate with no notice. First known instance of the US government forcing a deployed frontier model offline.·source

Peak-hour usagelow exposure

May 2026↑5h limits doubled; peak throttling reversed·source

Mar 2026↓Peak-hour throttling, ~7% of users hit session limits·source

Sustained sessionslow exposure

May 2026↑5h limits doubled; peak throttling reversed·source

Mar 2026↓Peak-hour throttling, ~7% of users hit session limits·source

Dec 2025↑Temporary holiday doubling·source

Aug 2025↓New weekly ceiling on top of 5h limits·source

High-frequency usagelow exposure

Jun 2026→Changes the reset logic for weekly usage limits from a rolling window based on session start to a fixed account-specific schedule.·source

Long-context workloadslow exposure

Jul 2026↑Enterprise tiers get a 500k context window on the default model (Team remains 200k), allowing larger document processing and longer session history.·source

Agentic / high-volume costlow exposure

Jul 2026↑New flagship Opus model at identical per-token pricing; Opus 4.8 still listed, so no forced migration or cost increase yet.·source

Jun 2026→New Sonnet 5 flagship at introductory $2/$10 per MTok — cheaper than Sonnet 4.6 until Aug 31, 2026, then reverts to the same $3/$15. Favorable near-term; no price increase.·source

Jun 2026→New premium Fable 5 tier at $10/$50 per 1M tokens — 2x Opus; Opus pricing unchanged·source

05 / WHAT CHANGED

Technical 5 · Product 7 · Business 3 · Policy 5

Mar 2026
↓ Peak-hour throttling, ~7% of users hit session limits
technicalImpact High · unannounced · source
Dec 2025
↑ Temporary holiday doubling
technicalImpact Low · source
Aug 2025
↓ New weekly ceiling on top of 5h limits
technicalImpact High · source

▸ View full constraint history (17 events)

Aug 2026
↓ Anthropic announced tightened biology-domain safeguards for Fable 5; announcement headline only, no published detail on which requests are affected.
policyImpact Low · source · secondary source
↔ shipped alongside New flagship Opus model at identical per-token pricing; Opus 4.8 still listed, so no forced migration or cost increase yet. (Jul 2026) — more for more, not a pure cut.
Jul 2026
↑ New flagship Opus model at identical per-token pricing; Opus 4.8 still listed, so no forced migration or cost increase yet.
productImpact Low · source · secondary source
Jul 2026
↑ Enterprise tiers get a 500k context window on the default model (Team remains 200k), allowing larger document processing and longer session history.
technicalImpact Low · source · secondary source
Jul 2026
↑ Reduces the minimum team size requirement for Claude Team plans, allowing smaller groups to access team-specific features.
policyImpact Low · source · secondary source
Jul 2026
→ Formalizes legal and data processing policies for K-12 educational usage.
policyImpact Low · source · secondary source
Jun 2026
→ New Sonnet 5 flagship at introductory $2/$10 per MTok — cheaper than Sonnet 4.6 until Aug 31, 2026, then reverts to the same $3/$15. Favorable near-term; no price increase.
businessImpact Low · source · secondary source
Jun 2026
↑ Restores access to Fable 5 model previously suspended due to export controls.
productImpact Low · source · secondary source
Jun 2026
↓ Anthropic's newest and most capable models became unavailable to every customer overnight by federal directive — users building on Fable 5 / Mythos 5 must migrate with no notice. First known instance of the US government forcing a deployed frontier model offline.
productImpact High · source
↔ shipped alongside Adds a new collaborative feature for teams. (Jun 2026) — more for more, not a pure cut.
Jun 2026
→ Changes the reset logic for weekly usage limits from a rolling window based on session start to a fixed account-specific schedule.
policyImpact Low · source · secondary source
Jun 2026
→ New premium Fable 5 tier at $10/$50 per 1M tokens — 2x Opus; Opus pricing unchanged
businessImpact Low · source
May 2026
↑ 5h limits doubled; peak throttling reversed
technicalImpact Medium · source
Mar 2026
↓ Peak-hour throttling, ~7% of users hit session limits
technicalImpact High · unannounced · source
Mar 2026
↑ Memory feature reached Free
policyImpact Low · source · secondary source
Dec 2025
↑ Temporary holiday doubling
technicalImpact Low · source
Sep 2025
→ Sonnet 4.5 released
productImpact Low · source
Aug 2025
↓ New weekly ceiling on top of 5h limits
technicalImpact High · source
Apr 2025
→ New Max tiers added
businessImpact Low · source

▸ Other product updates (3) — feature launches, not constraint changes

Jul 2026
↑ Adds access to a new model version for Pro users.
productImpact Low · source · secondary source
Jul 2026
→ Adds new specialized feature set for K-12 teachers.
productImpact Low · source · secondary source
Jun 2026
→ Adds a new collaborative feature for teams.
productImpact Low · source · secondary source