Greetings friends, There were a lot of model releases this week particularly from the Qwen Team at Alibaba, Also updated model from Suno, a stealth coding model 'code-supernova' which within the week itself updated its context window 5x. All this and more to read in this edition of 'This Week in All Things AI'
Readers who think others in their family, friends and acquaintances who have a curiosity in knowing more about rapidly evolving AI tools/services/use cases would benefit from being subscribers of this weekly newsletter are encouraged to share this publication link to them and invite them to subscribe
For those who use RSS readers to aggregate their reading of news such as the delightful NetNewsWire [MacOS,iOS, iPadOS] or Feedly [web, iOS, Android], this weekly newsletter is also available via the following RSS feed
https://paragraph.com/api/blogs/rss/%40twiata
Below is my personal website which aggregates links to many of my socials as well as content and community that I curate
goolamabbas | Instagram, TikTok | Linktree
Chief Knowledge Officer @ Animoca Brands. Also assists in Growth Marketing
The following messages were posted on the 'All Things AI' Telegram group from Sunday 21st Sep 2025 to Saturday 27th Sep 2025
For the coders in the group and using various agentic coding tools such as Cline, KiloCode, Cursor, Windsurf some links about a stealth model 'code-supernova' which has a 200K context window, multi-modal [supports image inputs] and which is free currently to use in these tools. Lots of speculation whether this is an upcoming model from Anthropic or xAI
>>
"code-supernova" is free to try, curious how you find it!
1 | Team Qwen from Alibaba on 🔥️️️️️️with 4 model releases in the last 24 hours. Simon Willinson with an excellent summary of their releases. I think the week has just begun for the Qwen team Four new releases from QwenIt's been an extremely busy day for team Qwen. Within the last 24 hours (all links to Twitter, which seems to be their preferred platform for these announcements): Qwen3-Next-80B-A3B-Instruct-FP8 and ...  |
2 | Poster: Roc Zacharias I’ve been wanting this for a while I mentioned it to Yusuf couple weeks ago, and now I found a team who is building. (Shiza.ai) We envision a decentralized future where individuals have ownership of their AI and are building the infrastructure and tooling for it: Empowering the Future: Why AI Ownership is the Next FrontierEmpowering the Future: Why AI Ownership is the Next Frontier By Syed Hussain As we stand on the precipice of a new era in technology, the discourse surrounding artificial intelligence often ...  |
3 | Beijing-based Kling AI, developed by Kuaishou Technology, released its 2.5 Turbo video model today, enabling creation of 1080p videos from text or image prompts with improved motion realism, camera control, and emotional depth. The update builds on the earlier 2.1 version by extending clip lengths up to 10 seconds, refining aesthetics to minimize flicker in complex scenes, and reducing pricing by 30% to enhance accessibility for filmmakers and creators. Early user demos on X showcase applications ranging from action sequences to nuanced character interactions. |
1 | OpenAI Rolls Out GPT-5-Codex to Developer APIs and IDE Tools OpenAI made its GPT-5-Codex model available through the Responses API, enabling integrations in tools like Cursor, GitHub Copilot, and Visual Studio Code. This specialized variant of GPT-5 features a 400,000-token context window for handling complex coding tasks such as refactoring and debugging, with built-in tool support for web search and Git operations. Developers have quickly adopted it for enhanced workflows, praising its adaptive reasoning and code review capabilities. https://x.com/i/trending/1970652477205680447
GPT-5-Codex is live now in the OpenRouter API!
The agentic coding variant of GPT-5 is purpose built for:  Real-world software engineering Adapts reasoning effort (fast for small edits, persistent on big refactors) Built-in code review to catch critical flaws
GPT-5-Codex is now available in Cursor.
Let us know your thoughts! |
2 | Team Qwen is definitely cooking and absolutely on 🔥️️️️️️with these new model releases The flagship model Qwen3-VL-235B-A22B is now open-sourced and available in both Instruct and Thinking versions: Qwen3-Max is here Upgrade of Qwen3-Coder, and the upgraded API `qwen3-coder-plus  We're thrilled to unveil Qwen3-VL — the most powerful vision-language model in the Qwen series yet! The flagship model Qwen3-VL-235B-A22B is now open-sourced and available in both Instruct and Thinking versions:  Instruct outperforms Gemini 2.5 Pro on key vision
 Qwen3-Max is here—no preview, just power!Qwen Chat:chat.qwen.ai Blog: qwen.ai/blog?id=241398… API: alibabacloud.com/help/en/model-… We’ve supercharged coding & agentic skills—now Qwen3-Max-Instruct without thinking rivaling top models on SWE-Bench, Tau2-Bench, 
We're excited to announce the upgrade of Qwen3-Coder, and the upgraded API `qwen3-coder-plus` is now available on Alibaba Cloud Model Studio with major improvements:
 Enhanced terminal task capabilities and better performance on Terminal Bench (w/ Qwen Code / Claude Code)
 |
3 | Suno, the Cambridge-based AI music startup, released its fifth-generation model v5 on September 23, 2025, available exclusively to Pro and Premier subscribers. The update introduces superior audio fidelity, authentic human-like vocals, and advanced tools for remixing and stem extraction, achieving an ELO score of 1293 in blind tests that surpasses previous versions and competitors. Early users on X have shared tracks demonstrating improved genre handling and production quality, while Suno continues to navigate copyright lawsuits from major labels. |
4 | Submissions for the Chroma Awards are now open. The Chroma Awards is the largest AI film, music and gaming competition with over $175,000 in cash prizes. Presented By Eleven Labs, Google Cloud, Freepik, Fal, Dreamina AI, Capcut |
5 | Comments from various members on Suno |
Research from NYU has found that frontier models from OpenAI, Google, and Anthropic can now pass all three levels of the CFA (chartered financial analyst) exam, including difficult Level III essay questions that eluded them two years ago.
New piece by @LesliePicker in @CNBC highlights the research led by @shilpinayak, our CTO at Goodfin, and @s_jagabathula in partnership with @NYUStern. The landmark paper finds AI can now pass the hardest level of the CFA exam — in a matter
https://www.cnbc.com/2025/09/24/ai-cfa-exam-pass-minutes-study.html
Shortcut positions itself as a superhuman on Excel.
For the spreadsheet nerds in the group, would love your opinion in case you give this a spin on its paid plan
I would assume free plans would be nerfed
While not perfect, Shortcut beats first year analysts from McKinsey/Goldman head-to-head 89.1% (220:27) when blindly judged by their managers.
We even gave humans 10x more time.
Try Shortcut now (before your boss does).
1) Agent accuracy is up 20%+ in last two weeks according to internal evals
2) Quick edits mode allows for easy multi-turn editing
3) Scaling well and errors are down ~75%
Oh, and MAX users can use Shortcut as an Excel plug-in too :)
The answer is Shortcut can do anything a human can on Excel.
Here I asked it to build a dinosaur themed dark-mode PMO template.
It's all in-distribution and it will all be solved.
1 | Factory AI recently launched an upgraded version of its Droid agentic coding tool which is causing the A-listers in AI in Cerebral Valley to rave about it when comparing it to Claude Code and OpenAI's Codex Short blurb from them
|
2 | The lab behind the code-supernova stealth model increased their context window to 1 million token from 200K tokens |
3 | Poster: Benjamin Chevalier Many of us here probably use MCPs, agents, “connectors”, etc… or even just AI-powered mail clients or whatnot. So I thought this could be relevant. It was discovered recently that a “postmark-mcp” quietly BCC’ed emails to an external address (https://www.koi.security/blog/postmark-mcp-npm-malicious-backdoor-email-theft). If you’re like me and like to experiment with local MCPs, keep this in mind. Using a different attack vector, last week someone managed to exfiltrate personal data through Notion’s agent using its web search tool. With Websearch, Notion agent can define URL and search query parameter. Attacker prompt injected a PDF the Notion agent ingested, asking the agent to gather personal data and ping the attacker’s server with the user data as search query, successfully exfiltrating personal data. Note that this could have come from a mail, a calendar invite, a github issue, anything https://www.codeintegrity.ai/blog/notion If you run a company, you’ll probably want to make sure employees with access to sensitive data are not installing random local MCPs.
Agentic systems with access to private data, ability to communicate externally and exposure to untrusted content essentially grants god‑mode, so verify sources, restrict sensitive access, and monitor to prevent silent exfiltration. |
I welcome feedback not only on this newsletter but across all the various surfaces where I curate content
The cover image of this newsletter was generated via Freepik using the Mystic 2.5 Flexible model
a sprawling underground city with intricately designed buildings on both the floor and ceiling, illuminated by cascading golden light from hidden sources
Regards, Yusuf