The Frontier

Qwen open-sources 27B model that runs on home GPUs

Editorial

Editorial

— 2 min read
Share
Qwen open-sources 27B model that runs on home GPUs

Two open-weights signals landed within hours of each other: Alibaba pushed the most practical member of its Qwen3.8 family out to the open-source community, and Meta's most expensive hire of the AI talent war walked out the door.

Qwen has open-sourced Qwen3.8-27B, a 27-billion-parameter dense multimodal model under Apache 2.0 that is small enough to run on consumer GPUs — free to download, deploy, and use commercially. The model, announced Friday, carries 262K tokens of native context (extendable to 1M with YaRN), and Qwen says it beats its own Qwen3.7-Plus on coding and office work. The genuinely new feature is reasoning_effort, which lets developers dial a model's thinking depth up or down per task so it doesn't burn tokens on easy questions. GGUF quantizations are already live for local runtime, and the release pushes Qwen's totals past 460 open-sourced models with more than 3 billion downloads. We dug into the strategy behind the flagship release this morning — Deep Dive — Alibaba's 2.4T open model is a strategy, not a gift — and the 27B is the other half of the same play: the giant MoE model is open to set the frontier, and now the consumer-class sibling is free to own the developer desktop. That combination is how an open-weight family captures mindshare before the paid tier ever gets a chance to compete.

Jiahui Yu, the multimodal lead of Meta's TBD Lab and the researcher reportedly hired for as much as $100 million a year in the 2025 talent war, announced this week that he is leaving Meta to start a new company. In a post on X, Yu — who previously co-led Gemini's multimodal work at Google Brain and ran OpenAI's Perception team through GPT-4o and GPT-4.1 — said building TBD Lab alongside Mark Zuckerberg and Meta's chief AI officer Alexandr Wang was "deeply inspiring," and that his team's Muse Spark, Voice Mode, Muse Image, and Muse Video work is something he's proud of. The exit lands days after Meta shipped Muse Spark 1.2, and it feeds a broader pattern: researcher-tracker alphaXiv counts more than 200 notable researchers who have left Meta, with 929 former Meta researchers on its roster. The new company's focus is undisclosed, but the arc is familiar — we mapped the broader trend in Why 37 AI-lab alumni founded startups in 2026. Meta paid mercenary prices to win the last talent war; the question is whether it can keep the missionaries.

What to watch: whether Qwen's reasoning_effort control becomes the template for efficient local models, and what Yu's mystery startup turns out to be.

If you could run a 27B multimodal model on your own hardware today, what would you build with it? Tell us in the comments.

Read more

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon turned its always-listening Bee line into a two-device family on Thursday, and NVIDIA opened a 2.8-trillion-token Indic corpus recipe. The connecting thread: both are bets that the next AI money is in memory — yours, or your model's. Amazon put two new always-listening devices on sale, and both now lead to a subscription. Bee Frontier, the next-generation wristband that also clips to clothing, ships now at $74.99 in Onyx, Lavender and Sage — up from the $50 predecessor — with seven-day