What DigitalOcean shipped in 2026

DigitalOcean Inference Router, Now Cache-Aware: Why the Cheapest Model Isn't Always the Best Deal

DigitalOcean’s Inference Router now supports cache-aware routing, preserving warm caches during agentic sessions to reduce costs and latency. The update introduces explicit model affinity headers and …

Safe RET Interrupt Vulnerability

AMD disclosed a potential vulnerability (AMD-SB-7061) where an attacker could inject an interrupt to weaken the Linux Safe RET mitigation for Speculative Return Stack Overflow (SRSO) on Zen 1–Zen 4 pr…

DigitalOcean replaces billing alerts with granular spend alerts

DigitalOcean introduced a new spend alert system that replaces legacy billing alerts, allowing users to set multiple percentage-based thresholds for monthly budgets across teams or organizations. The …

Under the Hood: Serving Kimi K3

DigitalOcean successfully launched Kimi K3 on day zero by optimizing hardware selection (NVIDIA HGX B300, AMD Instinct MI350X) and serving stack for a 2.78T-parameter model. They addressed dynamic too…

NVIDIA Nemotron

NVIDIA unveiled the Nemotron 3 family of open models with fully transparent training data, weights, and technical reports, offering hybrid Mamba-Transformer MoE architectures optimized for agentic AI.…

Kimi K3: Open Frontier Intelligence

Kimi introduced K3, a 2.8-trillion-parameter open model with native vision and a 1-million-token context window, marking the first open 3T-class model. It outperforms other tested models in long-horiz…

Outperforming Fable 5 at half the price: meet model synthesis, a new server-side tool on DigitalOcean Inference Engine

DigitalOcean introduced model synthesis, a server-side tool in its Inference Engine that orchestrates multiple open-source models in parallel to outperform single models like Fable 5 at roughly half t…

Upcoming GPU Pricing Updates

DigitalOcean will increase on-demand and 12-month reserved pricing for select NVIDIA and AMD GPUs starting August 1, 2026, citing strong demand and the need to expand high-performance compute access. …

Network Policies

Application-Specific Exclusions

Scale Faster with Managed Weaviate: Now in Public Preview on DigitalOcean

DigitalOcean introduced Managed Weaviate, a fully managed vector database service in public preview, starting at $20/month. The service handles operational tasks like backups, upgrades, and scaling wh…

DigitalOcean launches managed vector database service with full API suite

DigitalOcean introduced a new managed vector database service via its API, enabling users to provision, manage, and scale vector databases through endpoints like `/v2/vector-databases`. The service su…

DigitalOcean Evaluations: Production Model and Router Testing for the Inference Stack

DigitalOcean introduced Evaluations, a native feature in its Inference Engine, enabling teams to validate any model or inference router configuration on their own data before production. The tool supp…

Claude Claude Fable 5

Anthropic released Claude Fable 5 on July 1, 2026, restoring access after a June disruption. The model targets complex, long-running coding and knowledge work with state-of-the-art performance in codi…

Built for Mass Scale: Hard-Won Lessons from Teams Running High Volume Inference Workloads in Production

DigitalOcean hosted a panel at Deploy 2026 featuring Workato, Hippocratic AI, and ISMG, highlighting challenges in scaling AI inference workloads to production. Key issues included P99 latency spikes,…

DigitalOcean launches MCP tools for AI agent and model evaluations

DigitalOcean introduced a suite of MCP tools for its GenAI platform, enabling end-to-end evaluation of AI agents and raw models. The tools allow users to manage datasets, create test cases, run evalua…

deepseek-ai/DeepSeek-V4-Flashlike1.85kFollowDeepSeek139k

DeepSeek AI released DeepSeek-V4-Flash, a 284B-parameter Mixture-of-Experts model with 13B activated parameters, supporting a 1M-token context window. It uses FP4/FP8 mixed precision and achieves stro…

Run Codex in the cloud – DigitalOcean for Codex is now available

DigitalOcean introduced a Public Preview integration with Codex, enabling developers to provision persistent cloud development environments (Droplets) directly from Codex using natural language. The p…

Making sure you're not a bot!

DigitalOcean to charge 18% VAT in Peru starting June 2026

DigitalOcean will begin charging 18% VAT to customers with a tax location in Peru effective 1 June 2026, with charges appearing on invoices from 1 July 2026. Customers with a valid Peruvian RUC can av…