MON, OCTOBER 12, 2026
Independent · In‑Depth · Practitioner‑Tested
✎ AI News

Cloudflare Bought Deno — and Manus Raised $500M Without Meta

Cloudflare acquired Deno on 10 October with terms undisclosed and Node.js creator Ryan Dahl joining the Workers team, Manus closed over $500 million led by Tencent in its first round since China blocked Meta's acquisition, Nous Research confirmed a $1.5 billion valuation on a $90 million Series B, and OpenAI begins testing visual advertisements inside ChatGPT image generation in the US this month.

By AIToolsRecap October 12, 2026 8 min read 293 views
Home › Articles › AI News › AI News, 12 Oct 2026: Cloudflare Buys Deno, Man...

The five that matter

Cloudflare acquired Deno. Ryan Dahl, who created Node.js, is joining Cloudflare to work on Workers. Terms undisclosed.

Manus raised over $500 million. Led by Tencent, with Boyu Capital and IDG Capital. First round since China blocked Meta's acquisition.

Nous Research hit a $1.5 billion valuation. $90 million Series B led by Robot Ventures, with Nvidia, M12 and Samsung participating.

OpenAI is testing ads in ChatGPT. Visual advertisements shown during image generation, US only, this month.

Anthropic is giving startups a free year. Claude Team with up to five premium seats, plus $1,000 in API credits.

Cloudflare acquired Deno

Cloudflare announced on Friday 10 October that it is acquiring Deno, the company behind the JavaScript and TypeScript runtime of the same name. Financial terms were not disclosed. Deno had raised $26 million in total, including a Series A led by Sequoia.

Deno was co-founded by Bert Belder and Ryan Dahl. Dahl created Node.js in 2009 and then built Deno as a deliberate second attempt at the same problem, with the security model and TypeScript support he felt Node had got wrong.

The trigger appears to be celld, an open-source implementation of Cloudflare Workers that Deno released shortly before the deal. Kenton Varda, principal engineer for Cloudflare Workers, wrote that the Cloudflare team was "delighted" by it. Dahl said his team is "joining Cloudflare to make the Workers programming model a mainstream way to build servers."

The argument Cloudflare is making. Varda addressed the long-standing complaint that Workers was built differently from other cloud platforms to create lock-in. His response: "being open source and giving people an escape hatch is good business." Whether buying the company that built the escape hatch supports that argument or undermines it is the part developers are still arguing about.

Cloudflare has not said what happens to the Deno runtime or Deno Deploy. That is the question anyone with production code on either should be watching.

Full breakdown: Cloudflare acquires Deno — what it means for Workers.

Manus raised more than $500 million

Manus closed a round of over $500 million, led by Tencent, with Boyu Capital and IDG Capital participating alongside HSG. CNBC reported the round on 8 October.

The context matters more than the number. This is Manus's first raise since Chinese regulators blocked Meta's roughly $2 billion acquisition of the company in April. A startup whose exit was cancelled by its own government has now raised a quarter of that sum privately, at a moment when it was supposed to be in trouble.

Manus also released Manus 2.0 and is reported to be weighing a Hong Kong listing. We reviewed the product in full: What is Manus AI — full review, features and pricing.

Nous Research confirmed a $1.5 billion valuation

Nous Research raised a $90 million Series B led by Robot Ventures, with Nvidia, Microsoft's M12 fund and Samsung all participating. The round values the company at $1.5 billion.

The money is going toward turning Hermes, its open-source agent, into an enterprise product. Nous reports Hermes has been downloaded over 22.7 million times since February and claims it accounts for roughly 2.5% of global AI token usage. Business deployment is planned across Slack, Telegram and WhatsApp.

Treat the 2.5% figure as a company claim rather than a measured statistic — there is no independent audit of global token usage to check it against. The download count is more concrete.

OpenAI is putting ads inside ChatGPT

OpenAI will begin testing visual advertisements in ChatGPT in the United States later this month. The ads appear during image generation.

OpenAI says they will be labelled, kept visually separate from the generated image, and will not influence the model's answers. That last commitment is the one worth holding them to, because it is the only one that cannot be verified from the outside.

Why image generation first. It is the highest-intent surface ChatGPT has. Someone generating a product mockup, a logo or a room layout is demonstrably in a buying frame of mind, and the wait while the image renders is dead screen time that costs nothing to fill. If ads work anywhere in ChatGPT, they work there.

Anthropic expanded Claude for Startups

Announced at SF Tech Week, the expanded programme gives qualifying startups a free year of Claude Team with up to five premium seats, plus $1,000 in API credits.

Eligibility is either founding within the last five years or funding within the last two. That is a wider gate than most vendor credit programmes, which usually require an institutional round.

How to apply, what the credits actually buy, and where the limits bite: Claude for Startups — free Team year and $1,000 in credits.

Two open-weight models land this month

Both are worth a diary note rather than action today, because in both cases the weights are not out yet.

Mistral Large 4 went into public preview on the Mistral API on 6 October. Mistral's launch post gives 1 trillion parameters with 49 billion active per token and a 1M-token context window; the specification table on the same release lists 1.05 trillion and 52 billion, so treat the exact figures as unsettled. Preview pricing is $0.68 per million input tokens and $2.09 per million output, both at a 50% launch discount off list prices of $1.36 and $4.18. Open weights are due around 27 October. The licence has not been announced.

Reflection AI's Beam is a 501-billion-parameter mixture-of-experts model with 23 billion active per token and a 1M-token context window. Reflection reports 80.9 on SWE-bench Verified against 70.7 for Nemotron 3 Ultra, and claims three to four times less inference compute than comparable open models. Apache 2.0 weights are promised later in October. Access today is a waitlist.

On Terminal Bench v2.1, Beam scores 80.1 — below GLM-5.2 at 81.0, Kimi K3 at 88.3 and DeepSeek V4.1 Flash at 90.6. The efficiency claim is the interesting part, not the raw score.

Also worth knowing

  • Microsoft Work IQ began its October preview rollout, connecting Copilot and autonomous agents to live business data in Dynamics 365 and Power Platform. General availability and pricing are unannounced.
  • Satya Nadella called for an AI "emergency brake" — a mechanism letting authorised people pause or stop a model mid-task, with tamper-proof logs of significant model actions.
  • Apple's "Welcome Home" event is 13 October, expected to cover a home hub with a roughly six-inch display, a new Apple TV 4K and a refreshed HomePod mini. Apple is doing hands-on sessions rather than a keynote. Details are rumoured.
  • Global venture funding hit $159 billion in Q3 2026 per Crunchbase — down 25% from Q2 but up 53% year on year. A record 27 companies raised rounds of $1 billion or more, taking roughly a third of all venture capital.

What to do with today

  • If you run code on Deno Deploy, nothing breaks today, but Cloudflare has made no commitment about it. Know what your migration path looks like before you need it.
  • If you are an eligible startup, the Anthropic programme is the single most valuable thing on this page. Five seats and $1,000 in credits for a year is real money.
  • If you generate images in ChatGPT, expect advertisements within weeks if you are in the US.
  • If you self-host models, late October is busy. Mistral Large 4 weights around the 27th, Beam under Apache 2.0 at some point in the month.
Tags
AI News2026Generative AIAI agentsOpenAIAnthropic
⚑

Spot an inaccuracy?

We verify facts before publishing and correct errors promptly. If something in this article is wrong or outdated, let us know.

Report an error →