{"slug": "alibaba-ships-closed-weights-qwen4-max-and-puts-it-first-on-its-own-coding-chart", "title": "Alibaba ships closed-weights Qwen4-Max and puts it first on its own coding chart", "summary": "Alibaba's Qwen team released Qwen4-Max on Nov 16, the first model in its Qwen4 family, with a one-million-token context window and non-public weights; Alibaba's launch chart places Qwen4-Max first on the DeepSWE agentic-coding test ahead of the Anthropic, OpenAI and Moonshot (Kimi) models it charts, though it trails Anthropic on the CodeArena web-development ranking. Independent benchmarking firm Artificial Analysis scores Qwen4-Max 43 on its intelligence index against a median of 41 for similarly priced models, and lists pricing at $1.50 per million input tokens and $4.50 per million output, while calling the model \"very verbose.\" The model is available on Alibaba's QwenCloud and on Vercel's AI Gateway, which can route Claude Code, Codex or Cursor to it, and Alibaba says smaller Qwen4 models are coming to open source.", "body_md": "# Alibaba ships closed-weights Qwen4-Max and puts it first on its own coding chart\n\nAn independent benchmark puts it just above the median for its price, at $1.50 per million input tokens. Five more stories inside.\n\nToday leads with Alibaba. Its new model drew the weekend's biggest model thread on Hacker News, past 700 points. If you send coding work to cheaper models, it's one more to price out. Also today: GitHub retires six Copilot models on Dec 14. Claude Code Projects' coordinator now reports on its threads. A security firm found apps built with Lovable exposing user data without a login. A developer argues AI-written code should be open source. And Google says attackers ran an AI agent that turned security patches into exploits.\n\n## What changed this week\n\n### Alibaba ships closed-weights Qwen4-Max and puts it first on its own coding chart\n\nAlibaba's Qwen team released Qwen4-Max on Nov 16, the first model of its Qwen4 family, with a one-million-token context window. Its weights aren't public. Alibaba says smaller Qwen4 models are coming to open source. Its launch chart puts Qwen4-Max first on DeepSWE, an agentic-coding test, ahead of the Anthropic, OpenAI and Moonshot (Kimi) models it charts. It trails Anthropic's on CodeArena, a web-development ranking. Artificial Analysis, an independent benchmarking firm, scores it 43 on its intelligence index, against a median of 41 for similarly priced models. It lists $1.50 per million input tokens and $4.50 per million output. Artificial Analysis also calls the model \"very verbose.\" It's on Alibaba's QwenCloud and on Vercel's AI Gateway, which can route Claude Code, Codex or Cursor to it. Try it on one of your own tasks before trusting the chart.\n\n### GitHub Copilot drops six models on Dec 14, including Claude Fable 5 and Haiku 4.5\n\nGitHub said on Nov 13 that it will retire six models from every Copilot experience on Dec 14, \"including Copilot Chat, inline edits, ask and agent modes, and code completions.\" They are Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3. Each has a named replacement, such as Fable 5.2 for Fable 5, except Kimi K3: \"To be announced.\" Annual-plan individuals keep Haiku 4.5, GitHub says. Picked one by hand? Switch before Dec 14.\n\n### Claude Code Projects' coordinator now answers questions and reports on its threads, Anthropic says\n\nProjects, in beta in Claude Code, runs a stream of related tasks as parallel cloud sessions called threads. A \"long-running session where Claude acts as coordinator\" directs them. On Nov 13, Anthropic said on X, the coordinator began answering questions in the project chat and posting updates on running threads. Access is \"public beta on Pro, Max, and Ultra plans,\" not Team or Enterprise. It's not in the terminal. Each thread is a full session, so a project \"can reach usage limits faster.\"\n\n### A security firm finds thousands of Lovable apps' data tables open without a login\n\nLovable turns plain-language prompts into web apps. Crissy Field, a Hamburg security firm, checked 24,664 Lovable sites that load data from Supabase, a hosted database. On 7,403 of them, tables with email addresses answered read-only requests without a login. The firm says that \"should therefore not be read as a confirmed data leak rate.\" Its finds include \"more than 200,000 chat messages.\" The same scan found over 1,000 passport records. Lovable's terms put security on the customer, the firm notes. Built on Lovable? Check your access rules.\n\n### A developer says AI-generated code should be open source, in the weekend's top AI thread\n\nDeveloper Jakub Jankiewicz's short post says all code written with AI help should be open source or public domain under CC0, a no-rights-reserved license. He calls it a hot take. Models learned from open-source and Stack Overflow code, he argues, so closing their output \"feels like license laundering.\" Purely AI-generated code \"should not be copyrightable,\" he adds, citing the Monkey Selfie dispute, which \"established that non-humans cannot hold copyrights.\" Teams shipping closed code from agents will hear this argument.\n\n### Google says attackers ran an AI agent that turned security patches into working exploits\n\nGoogle's Threat Intelligence Group reported on Nov 16 that it has seen attackers \"fully automate vulnerability discovery and exploitation without a human in the loop for the first time.\" Their framework, dubbed exploit_agent, compares code before and after a patch, then writes and runs exploits for the gap. The report names no model behind it. It also describes stolen cloud credentials that \"generated millions of LLM API calls in approximately 24 hours.\" Patch quickly. Rotate any exposed API key.\n\n## Also worth your time\n\n- [llama.cpp and its ggml library reach v1.0.0, and breaking changes to their public API now bump the major version](https://huggingface.co/blog/ggml-org/llama-cpp-v1?ref=thenewway.ai)\n- [Claude Code on desktop takes custom themes through mods, Anthropic's plugin add-ons](https://code.claude.com/docs/en/plugins/mods/create?ref=thenewway.ai)\n\nKnow someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one.\n\n*The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.*", "url": "https://wpnews.pro/news/alibaba-ships-closed-weights-qwen4-max-and-puts-it-first-on-its-own-coding-chart", "canonical_source": "https://www.thenewway.ai/alibaba-ships-closed-weights-qwen4-max/", "published_at": "2026-11-16 15:28:27+00:00", "updated_at": "2026-11-16 15:52:05.732485+00:00", "lang": "en", "topics": ["large-language-models", "ai-products", "ai-tools", "generative-ai", "developer-tools"], "entities": ["Alibaba", "Qwen4-Max", "Qwen", "Anthropic", "OpenAI", "Moonshot", "Kimi", "Artificial Analysis"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/alibaba-ships-closed-weights-qwen4-max-and-puts-it-first-on-its-own-coding-chart", "markdown": "https://wpnews.pro/news/alibaba-ships-closed-weights-qwen4-max-and-puts-it-first-on-its-own-coding-chart.md", "text": "https://wpnews.pro/news/alibaba-ships-closed-weights-qwen4-max-and-puts-it-first-on-its-own-coding-chart.txt", "jsonld": "https://wpnews.pro/news/alibaba-ships-closed-weights-qwen4-max-and-puts-it-first-on-its-own-coding-chart.jsonld"}}