New: the first 100 founders get $120 in AI credits as Founding Members.Learn more
OSS ReleasesJun 16, 2026·2 min read·By StartupCorners

OSS Releases Week of June 16 2026: vLLM, Deno, Kubernetes

A busy week of OSS releases brings vLLM v0.23.0 with 408 commits, a feature-rich Deno v2.8.3, synchronized Kubernetes patch drops across four minor versions, and a security-conscious Astro bugfix.

Frameworks

  • withastro/astro astro@6.4.7, Fixes double URL-encoded paths that broke Sanity Studio and similar tools by replacing 400 rejection with iterative decoding, preserving the CVE-2025-66202 security fix; also corrects trailingSlash: 'never' locale URL helpers and slashes SSR bundle bloat from misrouted prerendered redirects in hybrid mode. Release

Infra

  • kubernetes/kubernetes v1.36.2, v1.35.6, v1.34.9, v1.33.13, Synchronized patch releases across four active minor versions; see each CHANGELOG for specifics. v1.36.2 | v1.35.6 | v1.34.9 | v1.33.13

AI

  • vllm-project/vllm v0.23.0, 408 commits from 200 contributors; DeepSeek-V4 sparse MLA decoupled from V3.2 and hardened across backends, Model Runner V2 now default for Llama and Mistral dense models, and the experimental Rust frontend gains streaming generate, dynamic LoRA, and new tool parsers. Release

  • BerriAI/litellm v1.89.1 (latest stable), Backports MCP, model-info, and DB fixes from the 1.84.8 patch set to the 1.89.x stable line; earlier stable branches (1.88.2, 1.87.3, 1.86.6, 1.85.5, 1.84.8) received the same coordinated backport wave covering Fable 5 support, batch-file auth, CrowdStrike AIDR, Mantle Responses SigV4, and DB-resilience fixes. Release

  • anthropics/anthropic-sdk-typescript sdk-v0.104.2, Removes retired models from the API and SDK surface. Release

  • langchain-ai/langchain langchain-openai==1.3.2, Adds correct GPT-4o image token counting; earlier point releases this week (langchain-core 1.4.5-1.4.7, langchain-anthropic 1.4.6, langchain 1.3.8-1.3.9) rolled up bug fixes, an allowed_prefixes feature for Anthropic, and a tornado security bump. Release

Runtime

  • denoland/deno v2.8.3, Feature-packed minor: deno compile gains watch mode, SubtleCrypto gets ML-DSA JWK import/export and the new SubtleCrypto.supports() static method, LSP gains doc-lint diagnostics and import-map hover info, OTEL trace sampler and span limits are now configurable, and node:http2 is auto-instrumented with OpenTelemetry. Release

  • cloudflare/workerd v1.20260616.1, Daily cadence releases across the week (June 11-16); changelog details are internal but each tag tracks the corresponding Workers runtime update. Release

Devtools

  • neovim/neovim v0.12.3, New stable patch release (now the stable tag); run :help news inside Nvim for the full changelog. Release

  • microsoft/vscode 1.124.2, Patch update to the 1.124 release; see the 1.124 release notes for full details. Release


The dominant theme this week is hardening at scale: vLLM lands a massive 408-commit release focused on production-grade DeepSeek-V4 and expanded Model Runner coverage, Kubernetes drops synchronized patches across four minor versions in a single pass, and LiteLLM pushes the same security and stability backports across six stable branches simultaneously, signaling that the AI infrastructure layer is maturing from rapid feature growth into disciplined, multi-track maintenance.

Ready for one link for your startup? Free to start, built from your website in seconds.

Founding Members get $120 in AI credits, a direct line to the team and a vote on the roadmap.

Free to start · No credit card · Work email, so we can read your website