> For the complete documentation index, see [llms.txt](https://docs.stg.ai.neevcloud.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.stg.ai.neevcloud.com/release-notes/may-2026-release.md). # May 2026 Release **Release Type:** Major Feature Release\ **Release Version:** v0.4.0 *** ## Overview The May 2026 release is our most feature-rich release to date. It introduces **Dedicated Model Deployment** and **CPU Virtual Machines** — both launching in **Beta** — alongside Bring Your Own AI Templates, scoped API Key Management, and significant billing improvements including free storage tiers and hourly-based network storage billing. This release also expands platform access through updated API documentation — advancing our mission to make NeevCloud a truly open, composable, and developer-first AI supercloud. *** ## What's New ### AI Inference — Dedicated Model Deployment *(Beta)* Users can now deploy and run NeevCloud validated GenAI models on dedicated GPU infrastructure through the NeevCloud AI platform. This release ships with **5 curated, ready-to-deploy models** spanning text generation, sentence embeddings, and text ranking. For the full list of supported models, hardware requirements, and pricing, refer to the [Dedicated Model Deployment documentation](https://docs.ai.neevcloud.com/ai-inference/). Each model can be deployed with a single click from the Model Catalogue. Once deployed, the model is accessible via a dedicated API endpoint scoped to your account. Billing is on-demand and charged hourly — usage lines appear shortly after the deployment ends and include the model name, GPU type, and replica count. **What's coming:** Playground integration for dedicated models, support for multimodal and image generation models, richer model detail views, improved GPU inventory and VRAM visibility, and faster model pull times via multi-tier caching are all on the near-term roadmap and will be addressed in upcoming releases. **Note:** Playground integration for dedicated models is not available in this release. Multimodal and image generation models are not yet supported through standard vLLM setups. These will be addressed in an upcoming release. ### Compute — CPU Virtual Machines *(Beta)* NeevCloud now supports CPU-based Virtual Machines (VMs), enabling teams to run traditional applications, Linux workloads, and non-GPU compute tasks directly on the platform alongside their AI infrastructure. VM instances can be provisioned and managed through the platform, with more OS images planned for upcoming releases. VM compute usage is billed hourly and appears in your usage history separately from VM disk storage, which follows the network storage billing model. **What's coming:** GPU VM support, additional OS images, finalised billing plan configurations for all VM types, and live migration capabilities are planned for upcoming releases. CPU VMs are the first step — the full VM service will evolve significantly in upcoming releases. **Note:** VM availability is subject to inventory and billing plan configuration currently being finalised for CPU VM configs. ### GPU Workloads — Bring Your Own AI Templates When deploying GPU workloads, you can now supply your own Docker templates — both public and private registry images are supported. This gives teams the flexibility to bring pre-built, customised environments without being limited to platform-managed templates. This is especially useful for teams with proprietary runtime dependencies, custom CUDA configurations, or specialised inference frameworks. Simply provide your template image URI during workload configuration to use it as the deployment base. ### Platform — API Key Management You can now create and manage scoped API keys for model and data-plane access directly from the platform. Each key is tied to specific resource types, so you can issue separate credentials for different services, team members, or automation pipelines — without sharing a single set of credentials across everything. Keys are tamper-resistant and auditable by design. This replaces ad-hoc credential sharing and lays the groundwork for enterprise-grade access control across the platform. If you have existing API keys, they remain valid and unaffected by this change. The new scoped key management gives you additional control to create, manage, and rotate credentials on a per-resource basis going forward. ### Platform — API Documentation Technical documentation for **AI Inference** and **AI Models** integrations is now publicly available for all customers. The updated API reference can be accessed at [docs.ai.neevcloud.com](https://docs.ai.neevcloud.com/). ### Billing — Free Storage Tiers & Hourly Network Storage Billing This release introduces meaningful billing changes that reduce upfront cost and improve predictability: * **100 GB of free persistent (network) storage** is now included across all accounts. * **25 GB of free local storage** is now included across all accounts. * Network storage billing has moved from an upfront deduction model to **hourly-based deduction**, meaning you are charged only for active usage rather than paying in advance. Usage history lines for storage are grouped monthly rather than hourly — you may not see a final usage line for a deleted volume until the month ends or shortly after deletion. ### AI Inference — Managed Template Upgrades Five AI Templates have been upgraded with the latest improvements and enhancements as part of the Managed Template Lifecycle Management (LCM) initiative. These upgrades are applied automatically — no action is required on your end. ### Platform — Resource Quota Management A centralised quota management system has been introduced internally as the single source of truth for resource limits across the NeevCloud platform. This ensures consistent enforcement of resource allocation policies across compute, storage, and model deployments. This change is infrastructure-level and requires no action from customers. *** ## Known Issues ### Dedicated Model Deployment * **Model detail view:** Clicking on a model card does not yet display full model metadata. A richer detailed view is planned for an upcoming release. * **GPU inventory and VRAM visibility:** When GPU inventory does not satisfy a model's VRAM requirements, the recommended behaviour and messaging are still being finalised. * **Model pull times:** Pull times for large models may be elevated in the current environment. Object storage caching is enabled; further improvements via multi-tier caching are being explored. * **"Last Used" not shown:** The "Last Used" timestamp is not currently displayed for Dedicated Inferencing API keys. * **List view:** The underlying LLM model name is not currently shown in the deployment list view. * **Playground not available:** Dedicated models cannot be tested in the Playground in this release. ### VM Service * **GPU VMs not yet available:** This release covers CPU VMs only. GPU VM support is planned for an upcoming release. ### Billing * **Usage line delay:** Usage lines for VMs and AI model deployments may appear with a short delay after the usage period ends. Network storage lines are grouped monthly and may not appear until month end. * **Currency:** NeevCloud billing is currently processed in USD. INR and other local currency amounts shown in the platform UI are approximate conversions. Full local currency billing support is planned for a future release. ### Storage * **Volume resizing not supported:** Storage volume size cannot be changed after creation. Create a new volume if you need additional capacity. *** ## Notes v0.4.0 marks a significant step in NeevCloud's evolution from an AI infrastructure platform to a full-stack AI supercloud. The addition of Dedicated Model Deployment, CPU VMs, BYO Templates, and scoped API Keys in a single release reflects the pace at which we are expanding the platform's capabilities. Both Beta features — Dedicated Model Deployment and CPU Virtual Machines — represent the beginning of their respective journeys on the platform, not the finished product. Each will receive continuous updates in upcoming releases, growing in stability, supported configurations, and functionality based on what we learn from early users. GPU VM support and Playground integration for dedicated models are already on the near-term roadmap and will ship in upcoming releases. As always, your feedback drives our roadmap. If you encounter issues or have suggestions, raise a support ticket or reach out to the NeevCloud support team or your account contact. *** ## Understanding Beta Features marked as *(Beta)* in this release are early-access capabilities that are functional and available to all users, but are still under active development and continuous improvement. Here is what Beta means in the context of NeevCloud: * **Early access, not yet recommended for critical production use.** The core functionality works, but the feature has not reached the stability, completeness, and performance standards required for General Availability (GA). We recommend against relying on Beta features for time-sensitive or business-critical production workloads at this stage. * **Under continuous improvement.** Beta features are actively being developed. Expect frequent updates to functionality, reliability, performance, and UX in upcoming releases. * **Known gaps exist.** Beta features may have UI flows that are not yet fully complete, or certain views may not yet show all available information. Known gaps for each feature are documented in the Known Issues section above. * **APIs and interfaces may evolve.** While core endpoints are stable, some parameters or response formats may be refined before a feature reaches GA. * **No full SLA guarantees.** Beta features are not covered by the same uptime and reliability commitments as GA features. * **Your feedback shapes the roadmap.** Beta is a collaborative phase. We actively use feedback from early users to decide what gets built, fixed, and improved next. Raise a support ticket or reach out to your account contact to share what you experience. Beta features will graduate to GA once they meet our internal quality, performance, and completeness thresholds. We will announce GA graduation clearly in the release notes.