AILIMIT.WATCH
LIVE

Grok

Last verified: Jun 6, 2026

1 unannounced

19 sourced changes·47% 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: loosening · drift +15% over 90d · 1 tightening in 90d · 1 silent change in 90d

Constraint Stability

100/100

higher = fewer recent restrictions

loosening+15% / 90d
1 unannounced change

Grok has eased its constraints recently.

Most noticeable: Image-generation policy tightened after regulatory pressure

Recency
65
days since it last tightened
Velocity
2.6
tightenings per year
Silent
33%
of tightenings were unannounced
Reversals
0%
of tightenings later reversed (same dimension)

Grok constraint metrics — computed from 3 tracked tightenings · median 180 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

Grok churns fast on nearly every axis. Model access loosens relentlessly: Grok 4.1 shipped free to all tiers in November 2025, the multi-agent Grok 4.20 beta followed in February 2026, then Grok 4.3 (native video, 1M context) in April, and Grok 4.5 landed on the xAI API in July 2026 at $2/1M input and $6/1M output. The top of the range is steep — the SuperGrok Heavy tier launched at $300/mo alongside Grok 4. Policy is what complicates the picture. The July 2025 'MechaHitler' incident left an unsafe system-prompt path live roughly 16 hours before rollback, and in January 2026 a regulator-driven overhaul (UK ICO, France, California AG) banned sexual images of real people after a permissive 'Spicy Mode' era. Developers should also note a quiet reversal: priority processing (service_tier) was added across text, image, and video endpoints in June 2026, then silently narrowed to text-only in July — a tightening flagged as silent in the record. Capability and access mostly expand, but the surrounding rules — safety policy, priority scheduling — can shift abruptly, sometimes without announcement and sometimes under external legal pressure. Model availability and features arrive quickly and generously on Grok; the terms around them are not stable commitments. Budget for both the $300 ceiling and the chance that a feature you depend on is scoped down between releases.

03 / CURRENT LIMITS

04 / WHERE IT MATTERS

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

Peak-hour usagelow exposure
Jul 2026↓Priority processing is no longer available for image and video inference endpoints, restricting high-priority scheduling to text inference only.·source
Jun 2026↑Allows users to opt-in to higher scheduling priority for inference requests, potentially reducing latency at an additional cost.·source
Sustained sessionslow exposure
Jun 2026↑Enables long-running autonomous tasks in Grok Build.·source
Agentic / high-volume costlow exposure
Jul 2026→Adds new automation capabilities to the Grok platform.·source
Jul 2026↑New model availability expands capabilities for coding and agentic tasks.·source
Jul 2026→Establishes pricing structure for the new Grok 4.5 model.·source
Specific-model dependencelow exposure
Jul 2026↑New speech-to-speech model available, and the grok-voice-latest alias silently repoints to it on August 5, 2026.·source

05 / WHAT CHANGED

Technical 0 · Product 15 · Business 2 · Policy 2

  1. Jan 2026

    ↓ Image-generation policy tightened after regulatory pressure

    policyImpact Medium · source · secondary source
  2. Jul 2025

    → New $300/mo Heavy tier

    businessImpact Low · source · secondary source
  3. Jul 2025

    ↓ System-prompt safety incident

    policyImpact Medium · source
▸ View full constraint history (10 events)
  1. Jul 2026

    ↑ New speech-to-speech model available, and the grok-voice-latest alias silently repoints to it on August 5, 2026.

    productImpact Low · source · secondary source
  2. Jul 2026

    ↑ Increases accessibility to Grok Build by open-sourcing the project.

    productImpact Medium · source · secondary source
  3. Jul 2026

    → Establishes pricing structure for the new Grok 4.5 model.

    businessImpact Low · source · secondary source
  4. Jul 2026

    ↓ Priority processing is no longer available for image and video inference endpoints, restricting high-priority scheduling to text inference only.

    productImpact Low · unannounced · source · secondary source

    ↔ shipped alongside New model availability expands capabilities for coding and agentic tasks. (Jul 2026) — more for more, not a pure cut.

  5. Apr 2026

    → Grok 4.3 released

    productImpact Low · source · unconfirmed
  6. Feb 2026

    → Multi-agent Grok 4.20 beta available

    productImpact Low · source · secondary source
  7. Jan 2026

    ↓ Image-generation policy tightened after regulatory pressure

    policyImpact Medium · source · secondary source
  8. Nov 2025

    ↑ Grok 4.1 available free to all tiers

    productImpact Low · source
  9. Jul 2025

    → New $300/mo Heavy tier

    businessImpact Low · source · secondary source
  10. Jul 2025

    ↓ System-prompt safety incident

    policyImpact Medium · source
▸ Other product updates (9) — feature launches, not constraint changes
  1. Jul 2026

    ↑ Adds new video generation capabilities including text-to-video and image-to-video.

    productImpact Low · source · secondary source
  2. Jul 2026

    ↑ Adds granular control over audio processing, reducing potential for missed speech in noisy environments.

    productImpact Low · source · secondary source
  3. Jul 2026

    ↑ Expands model access to users in the EU region.

    productImpact Low · source · secondary source
  4. Jul 2026

    → Adds new automation capabilities to the Grok platform.

    productImpact Low · source · secondary source
  5. Jul 2026

    ↑ New model availability expands capabilities for coding and agentic tasks.

    productImpact Low · source · secondary source
  6. Jun 2026

    ↑ Enables long-running autonomous tasks in Grok Build.

    productImpact Low · source · secondary source
  7. Jun 2026

    ↑ Allows users to opt-in to higher scheduling priority for inference requests, potentially reducing latency at an additional cost.

    productImpact Low · source · secondary source
  8. Jun 2026

    ↑ Adds new capabilities for file storage, public URL generation, and direct integration with Imagine endpoints.

    productImpact Low · source · secondary source
  9. Jun 2026

    ↑ Adds new model access for developers via API.

    productImpact Low · source · secondary source

06 / OUTLOOK

Volatile

Volatile. Last 12 months: 1 pricing, 0 limit/quota, 15 model-availability changes (2 tightenings).

Observed past cadence — not a prediction.

07 / USER SIGNALS

USER SIGNALS — HACKER NEWS · GITHUB · APP STORE

17 complaint mentions · last 180d

9 HN · 8 App Store

  • Pricing·847%
  • Limits / quota·529%
  • Performance·212%
  • Model removal·16%
  • Other·16%

Hacker News, GitHub issues & Apple App Store reviews matching complaint keywords — not a scientific sample. See methodology.

08 / ALTERNATIVES

ALTERNATIVES

By constraint history — not capability.

More stable:Gemini, v0, Bolt.new
Similar risk:Claude, Devin, Lovable
→ Full comparison matrix

09 / RELATED