Hi,
This is the first VibeStack weekly. Each week you get the AI releases we logged, with a source for every one, plus any ranking changes. It takes about two minutes to read.
New this week
- Reflection Beam (16 Oct): Reflection's first open-weight model. It is a sparse mixture-of-experts model with 501B total parameters and 23B active, built for coding, reasoning and agentic work. The weights are on Hugging Face under Apache 2.0, and the model card lists a 1M-token context window. → https://vibestack.tools/tools/beam/
- Meta Muse Spark Open (15 Oct): open weights. A natively multimodal MoE model with 816B total parameters, 42B active and a 1M-token context, aimed at coding and agentic work. Meta says it runs on two M5 Ultra Mac Studios with MLX. The license is the Muse Model License, which has limited commercial terms. → https://vibestack.tools/tools/muse-spark-open/
- Gemini 4 Argon (13 Oct): now available to developers. It is in the API (AI Studio and the Gemini platform) and in the Gemini app for AI Ultra and Pro subscribers. Google's API docs list introductory pricing of $2 per 1M input tokens and $10 per 1M output tokens until 30 Nov, rising to $4 and $20 after that. It is not on the free tier. → https://vibestack.tools/tools/gemini/
- Lovable (15 Oct): apps you build in Lovable can now use AI models and tools. The list includes GPT-Live, the latest Claude models, Gemini 4 Argon, Perplexity, Firecrawl and more. You add them by describing what you want in the project chat. → https://vibestack.tools/tools/lovable/
- Figma at Config India (15 Oct). Code Layers and Figma Specs are in open beta, alongside Payload scaffolding in Figma Make, Guidelines for agents, and MCP improvements. → https://vibestack.tools/tools/figma/
- Cursor's Origin App Marketplace (13 Oct). It launched with 20+ apps for deploys, CI, code review, security and monitoring, and public submissions are open. → https://vibestack.tools/tools/cursor/
- Hermes Agent v0.22.0 (13 Oct). Adds guided desktop onboarding on a free trial model, app connectors, one-click local llama.cpp models and a 600-plugin catalog. → https://vibestack.tools/tools/hermes-agent/
- OpenUI Lang 1.0 (12 Oct). Thesys's MIT-licensed language lets a model stream UI built from your own component library. The 1.0 features need the React renderer for now. → https://vibestack.tools/tools/openui/
Also recent: Synthesia launched Syren in beta (9 Oct), which turns a prompt into motion graphics, avatars and voiceover. Google introduced Playground for prompt-made games (7 Oct, US 18+). Anthropic shipped Claude Haiku 5.5 (7 Oct). All releases with sources: https://vibestack.tools/releases/
Top picks: Best AI coding agents
- Claude Code (4.7): the default agent when the task is a large existing codebase and you're happy in the terminal or an IDE plugin.
- Cursor (4.7): the strongest daily driver if you want the agent and the editor to be the same place.
- Codex (4.4): OpenAI's repo agent, and the one to try if your models and account already live there.
- Devin (4.3): the cloud agent you hand a task and check on later.
Also ranked: Cline, Aider, GitHub Copilot and Devin Desktop. The full shortlist, with a reason for every pick: https://vibestack.tools/rankings/coding-agents/
Ranking highlight: Cursor vs Claude Code is a dead heat
Both score 4.7. Pick Cursor if you want the agent inside your editor. Pick Claude Code for a big repo from the terminal or an IDE plugin. The verdict, the comparison table and sources: https://vibestack.tools/compare/cursor-vs-claude-code/
Building without a local setup? See Lovable vs Bolt vs v0: https://vibestack.tools/compare/lovable-vs-bolt-vs-v0/
Know someone who ships with AI?
Get this in your inbox every week. Got a tool we should review? https://vibestack.tools/submit/
Disclosure: some tool pages on VibeStack include affiliate (partner) links, and these are labelled. If you buy through one, we may earn a commission at no extra cost to you. Featured listings are paid and labelled too. Neither one ever changes a score or a ranking. Details: https://vibestack.tools/disclosure/