What shipped · 2026-w42
What shipped in AI dev tools: week of Oct 12, 2026
Oct 12, 2026 – Oct 18, 2026
16 releases across 9 tools this week; biggest: Vercel AI SDK (5 releases).
Vercel AI SDK
DeepSeek V4.1 Flash 1017 on AI Gateway
On October 17, 2026, DeepSeek V4.1 Flash 1017 became available on Vercel AI Gateway as deepseek/deepseek-v4.1-flash-1017. Vercel describes it as a minor update to DeepSeek V4.1 Flash that improves reasoning, coding, agentic tasks, and image understanding, and that finishes tasks faster with shorter outputs than the previous snapshot.
Source for DeepSeek V4.1 Flash 1017 on AI Gateway (opens in a new tab)Elorian Glimpse on AI Gateway
On October 16, 2026, Elorian's Glimpse visual reasoning model became available on Vercel AI Gateway as elorian/glimpse. It accepts images and video to follow objects through motion, interpret spatial relationships, and connect an action to its physical consequences. The AI SDK example sends an image or, with AI SDK 7, a video file part.
Source for Elorian Glimpse on AI Gateway (opens in a new tab)Perplexity Decider V1.2 on AI Gateway
On October 16, 2026, Perplexity Decider V1.2 27B (perplexity/pplx-decider-v1.2) and Decider Flash V1.2 4B (perplexity/pplx-decider-flash-v1.2) became available on Vercel AI Gateway. Both are multimodal decision models for classifying content, routing requests, and scoring text, structured records, and images. Boolean questions return a likelihood, Choice questions select from categories you define, and Score questions use an ordered rubric.
Source for Perplexity Decider V1.2 on AI Gateway (opens in a new tab)Reflection Beam on AI Gateway
On October 16, 2026, Reflection AI's Beam, previously in stealth as stealth/cat-jam, became available on Vercel AI Gateway. Beam is an open-weight reasoning model for coding and agentic tasks, released under Apache 2.0, with tool calling, structured outputs, and a 1M-token context window. It activates 23B of 501B parameters per token. During the launch period it is free when serviced by Reflection AI, as reflection/beam or reflection/beam-free.
Source for Reflection Beam on AI Gateway (opens in a new tab)Muse Spark Open on AI Gateway
On October 15, 2026 Meta released Muse Spark Open, an open-weight multimodal model for coding and agentic work with 816B total parameters, 42B active, and a 1M-token context window. The same day Vercel AI Gateway made it available as meta/muse-spark-open, with text and image input, through the AI SDK and in coding agents including Claude Code, Codex, and Cursor.
Source for Muse Spark Open on AI Gateway (opens in a new tab)
Vercel
AI Gateway logs show session affinity
On October 18, 2026, Vercel AI Gateway logs gained a Sessions view. It groups requests into sessions and shows how session affinity routed each one, including the coding agent or app, cost, models, providers, and how much of the input was cached. Session affinity runs by default on text and chat requests. To group requests yourself, send the same sessionId in providerOptions.gateway, or the same x-session-id header, on every request in a conversation.
Source for AI Gateway logs show session affinity (opens in a new tab)Python runtime generally available
On October 15, 2026, Python became a generally available runtime on Vercel. FastAPI, Flask, and Django apps are recognized automatically, CPython runs on Linux with uv and pyproject.toml, and request-driven apps can use Fluid compute while long-running pull-based work can use Vercel Daemons. Python SDKs in the same release include Sandbox, an AI SDK beta, a Workflow SDK beta, Queues (beta), Connect, and Blob.
Source for Python runtime generally available (opens in a new tab)Vercel Daemons public beta
On October 15, 2026, Vercel Daemons entered public beta. Daemons run containers on Fluid compute, deployed with the app, that pull work from Vercel Queues, databases, or third-party systems without an HTTP server. That includes queue consumers (Vercel Queues poll mode, Celery, BullMQ, RabbitMQ, or Kafka), Temporal or Cadence workers, and self-hosted CI runners for GitHub Actions, GitLab, or CircleCI.
Source for Vercel Daemons public beta (opens in a new tab)
Claude Code
Claude Code 2.1.302
Released October 16, 2026. Adds CLAUDE_CODE_AUTO_MODE_MODEL to choose the model the client-side auto mode classifier uses, an Idle compaction row in /config, and a notice that standalone Claude Design (claude.ai/design) closes by December 14, 2026. Image reads over 128 MB are refused, plugin startup does less file-system work, and the release includes a large fix sweep for cloud sessions, MCP, and worktrees.
Source for Claude Code 2.1.302 (opens in a new tab)Claude Code 2.1.300
Released October 14, 2026. Adds mod prompt virtualText decorations, agent transcript view via $.ui.view({ agentId }), /advisor on Claude Platform on AWS, richer OpenTelemetry GenAI attributes, and a large stability sweep for session resume, worktrees, and cloud sessions. Source: official Claude Code changelog.
Source for Claude Code 2.1.300 (opens in a new tab)
Cursor
Cloud agent canvases and environment improvements
On October 16, 2026, cloud agents can build interactive canvases and share a finished canvas by link. The Environments API is in beta for creating and managing the cloud environments agents run in. Cloud agent environments support up to 300 repositories, and large environments load 5x faster.
Source for Cloud agent canvases and environment improvements (opens in a new tab)
Figma
Figma Specs and agent MCP updates
On October 15, 2026 Figma launched Specs (a markdown editor for design language/context for agents), moved code layers into public beta, and expanded Figma MCP so agents can generate code layers, work with specs, create visual assets, and run published Weave workflows from tools like Cursor and OpenCode.
Source for Figma Specs and agent MCP updates (opens in a new tab)
GitHub Copilot
Agentic autofix for code scanning alerts
On October 15, 2026, agentic autofix for code scanning alerts became generally available. Assigning an alert to Copilot explores relevant files, proposes a fix, reruns CodeQL to confirm the alert is closed, and opens a draft pull request for review. It is available to GitHub Code Security and GitHub Advanced Security customers on GitHub Enterprise Cloud once an enterprise admin enables the Agentic autofix policy and a Copilot billing method is defined.
Source for Agentic autofix for code scanning alerts (opens in a new tab)
Lovable
AI models and tools in Lovable apps
On October 15, 2026, apps built in Lovable can use new AI models and tools without leaving the builder: GPT-Live, the latest Claude models, Gemini 4 Argon, Gemini Omni, and Lyria, plus Perplexity, Firecrawl, Parallel, and Jev by Typesafe AI. Name one in chat, or let Lovable choose a model or tool for the feature you describe.
Source for AI models and tools in Lovable apps (opens in a new tab)
Replit
Bring Your Own Cloud
On October 16, 2026, Replit introduced Bring Your Own Cloud for Microsoft Azure, Google Cloud, or Amazon Web Services. Full BYOC runs the Replit platform and published project hosting in the customer's cloud tenant and is a private preview for enterprise customers. BYOC for Published Apps hosts published projects in that tenant, and Image Exports provide a Docker image, a Terraform file, and an SBOM so a team can self-host. Both of those are in early access for enterprise customers.
Source for Bring Your Own Cloud (opens in a new tab)
v0
Code tab downloads and iOS pickers
On October 16, 2026, large files that cannot be previewed can be downloaded from the v0 Code tab, which shows a message and keeps a Download action in the file header. The v0 iOS app adds a searchable skills picker, a design system picker with the same examples as the web, and video attachments you can preview with Quick Look.
Source for Code tab downloads and iOS pickers (opens in a new tab)
Updated