Grok Retired Models List: What Stopped Working May 15, 2026
May 15, 2026 marked a significant breakpoint in the AI landscape, especially for users of Grok models developed by SpaceXAI and supported by xAI. Several Grok models were officially retired on this date, triggering shifts in pricing, API billing structures, and user tier distinctions that have confounded some while streamlining options for others. This post offers a detailed rundown of what happened, the pricing implications, and what to expect moving forward.
Timeline of Grok Model Retirements in 2026
Grok models have long been a staple in the AI tooling space, especially for developers and small teams leveraging B2B SaaS AI products. However, May 15, 2026 serves as a key "breakpoint" date when several versions were phased out:
- Grok 3 retired: Complete cessation of support and API calls
- Grok 4 fast retired: No longer functional as of May 15
- Grok 4.1 fast retired: Also disabled on the same day
In short, users relying on these specific Grok versions needed to migrate or upgrade well before this date. The forced retirements were part of a strategic move by SpaceXAI and xAI to focus on newer, more efficient, and commercially optimized AI models bundled under the broader 'X-bundled' tiers.
What Caused These Retirements?
The core driver was a mix of cost rationalization and technological evolution. Grok’s architecture and token handling hadn’t kept pace with recent advances, leading to inefficiencies in API per-token billing and increased compute costs. The move away from older Grok versions also aligns with xAI’s push to standardize billing with cached input tokens pricing and to consolidate tiers for clearer consumer engagement.
API Per-Token Billing vs Cached Input Tokens Pricing
One notable pain point prior to May 2026 was the billing confusion stemming from API per-token pricing, which charged users strictly based on every token processed in requests and responses. This was often opaque for end-users, especially those using Grok versions with poor token optimization.
Post-retirement, xAI and SpaceXAI pivoted toward a cached input tokens pricing scheme for newer models, which caches aibusinessweekly.net frequent inputs and drastically cuts cost for repeat queries. This was part of the rationale to retire older variants like Grok 3 and Grok 4 fast versions.
Consumer Tiers vs X-Bundled Tiers: Pricing and Access Changes
Another crucial change accompanying the model retirements was restructuring of user tiers:
- Consumer tiers: These are primarily intended for small teams, hobbyists, and educational users. They retain limited but affordable access to AI functionalities, often at a slightly reduced capability and strict token quotas.
- X-bundled tiers: Aimed at enterprises and professional developers, these tiers bundle multiple AI services with optimized performance and newer model access, priced higher but with fewer limits.
For example, SuperGrok Lite remains one of the few affordable offers tailored for smaller teams. It’s priced at $10/month or $100/year (effective $8.33/month), reflecting the standard annual billing discount that encourages long-term commitments.
Annual Billing Discount Math Explained
Annual billing discounts typically offer a 16.7% saving compared to monthly payments. Using SuperGrok Lite as an example:
Billing Option Total Annual Cost Effective Monthly Cost Discount vs Monthly Monthly ($10/month) $120 $10.00 — Annual ($100/year) $100 $8.33 16.7% off monthlyWhile the math is simple, it’s critical not to round prices prematurely on pricing pages or documentation—the source always uses exact numbers, and small rounding errors can mislead budget calculations for teams.
Free Tier Limitations and Known Unknowns
Following the May 15, 2026 breakpoint, the free tiers provided by SpaceXAI and xAI for Grok access have become substantially more limited. With the retirement of older models, free tier users lost access to Grok 3, Grok 4 fast, and Grok 4.1 fast.
What remains unclear includes:
- Whether cached input tokens pricing will apply uniformly to free tiers
- How strict the token and input limits will become with future Grok replacements
- Potential throttling or API call limits imposed due to infrastructure costs
Error-prone or opaque billing pages have historically complicated independent verification of free tier capabilities. This remains a grievance for many users eager for transparent pricing and functionality disclosures, especially in the wake of these retirements.
Implications for Small Teams and Developers
Small teams and developers who relied on older Grok models must evaluate the costs and benefits of migrating to newer models under the X-bundled tiers or accept the reduced access in consumer tiers. SuperGrok Lite is a notable budget-friendly option, but it has capped token usage reflecting typical consumer tier constraints.
API users should pay careful attention to token pricing changes, especially with the shift towards cached input tokens. This will affect monthly cost predictability and budgeting in ways that weren’t as evident before May 15, 2026.
Summary of Key Dates and Models Retired
Model Status as of May 15, 2026 Notes Grok 3 Retired No more API support or billing Grok 4 fast Retired Discontinued for efficiency and tech updates Grok 4.1 fast Retired Also stopped on May 15, 2026Final Thoughts
The May 15, 2026 retirements provide a critical breakpoint for both vendors like SpaceXAI/xAI and end-users of Grok AI solutions. While the retirements render some legacy investments obsolete, they push the industry towards leaner, more sustainable pricing models and better-performing AI tiers. Small teams and developers should recheck their dependencies and plan for updated pricing strategies.
For those leveraging Grok and related models, awareness and adaptation to API per-token billing changes, cached input optimizations, and tier restructurings like SuperGrok Lite’s pricing will be essential. Transparency on free tier limits, ongoing price changes, and feature shifts must be monitored carefully as 2026 progresses.