The Frontier

Nikkei's AI scorecard puts Kimi K3 sixth — and Meta poaches another OpenAI researcher

Editorial

Editorial

— 2 min read
Share
Nikkei's AI scorecard puts Kimi K3 sixth — and Meta poaches another OpenAI researcher

Two signals about where AI's center of gravity sits right now: a Japanese survey says the strongest open model is now beating closed giants on trust — and the talent war for the people who build them keeps escalating.

Nikkei's joint survey with Weights & Biases has Kimi K3 climbing to sixth overall, scoring ahead of Google's and xAI's flagship models. The "AI Model Score" comparison, run by Nikkei Digital Governance with the developer-tools company Weights & Biases, ranks Moonshot AI's Kimi K3 sixth in its latest table — with the Chinese open-weights model scoring high specifically on reliability measures, per the report first flagged by Chinese-language feeds this week. The detail that should sting US labs: an open model built outside the frontier club is out-ranking models backed by some of the largest compute budgets on earth.

The ranking is less a trophy than a purchasing signal. Kimi K3 already anchors production tooling — legal AI company Harvey post-trained it as the base for its long-horizon legal agent, and we covered that adoption last week — and reliability scores are exactly what enterprise buyers weigh before routing real workloads through a cheaper model. That is the same dynamic reshaping Anthropic's business: corporate spending is drifting from priciest flagship to cheaper alternatives, and every independent scorecard that legitimizes open weights accelerates the drift. When the model you can self-host starts beating closed giants on trust, "frontier" stops being the default purchase.

Meta's Superintelligence Labs keeps buying OpenAI muscle: researcher Luke Metz has jumped ship, according to Axios. Metz — a veteran of the original ChatGPT-era research teams who left OpenAI for startup TML earlier this year, then returned to OpenAI at the start of 2026 — will report to Alexandr Wang, the former Scale AI chief running Meta's superintelligence push. Axios attributes the move to a source familiar with the matter; neither lab has issued a formal announcement, so treat the reporting as confirmed-by-sourcing rather than press-released.

One hire is not a strategy, but the pattern is the story: Meta has spent 2026 systematically pulling researchers out of OpenAI, from ChatGPT co-creator Shengjia Zhao down the roster, pairing poached talent with Wang's operational control and a compute budget few startups can match. For OpenAI the exits compound a harder quarter — pricing pressure from cheap open models, and rivals hiring away the people who know how its systems are trained. The talent market is doing what benchmarks only suggest: repricing where the frontier actually lives.

What to watch: whether Kimi K3's enterprise traction holds after the Harvey-style integrations mature — and whether OpenAI answers Meta's raid with retention packages of its own.

Open models winning on reliability while labs fight over the humans who build them — which shift matters more over the next year? Tell us in the comments.

Read more

Houmo moves compute inside memory — and A20 enters production

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

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