The Frontier

Bilibili opens a 100-model arena — GPT-6 Astra takes the top slot

Two stories where the ranking comes from outside the lab: China's biggest video platform just turned creator reviews into a live model leaderboard, and a clinical voice company published the first benchmark of whether AI can actually say a drug's name. Bilibili's new AI Infinite Arena ranks models by how often they win — not by how well they score. The platform opened a leaderboard fed by its UP hosts (creators), who test more than a hundred models on their own terms: code, reasoning, collabora

The Arena

Huawei Cloud will sell its agent platform overseas from December 30

Huawei Cloud used its annual conference to put a date on taking enterprise agents global — and to argue that the open-source edition is doing the selling. Huawei Cloud will make AgentArts, its commercial enterprise agent platform, available in markets outside China on December 30, the company said at HUAWEI CONNECT 2026 in Shanghai. AgentArts ships alongside openJiuwen, an open-source edition of the same platform, and Huawei says the pair together expose more than 5,000 general MCP assets plus

The Arena

CXMT's fifth-gen DRAM platform enters mass production

China's largest memory maker says its new process platform is now in volume production, and the first products built on it are already going into domestic flagship phones. ChangXin Memory Technologies said Sunday that its fifth-generation technology platform has entered mass production, and that 24-gigabit LPDDR5X products built on it are already shipping. The announcement was made at the 2026 World Manufacturing Conference in China. CXMT puts the platform's memory-array active-area half-pitch

The Guardrails

Paid AI subscribers sue four labs over their slowdown pledge

Four people who pay for ChatGPT, Claude, Grok and Gemini are suing the companies that sell them, arguing that the labs' public push to "pace" frontier development is an illegal agreement to restrain trade. The class action — Buist v. Anthropic PBC, No. 3:26-cv-10693 — was filed Friday in the U.S. District Court for the Northern District of California and names Anthropic, OpenAI, SpaceXAI and Google. The plaintiffs are four consumers who hold paid subscriptions to the four services, and they are

The Frontier

ProgramAsWeights compiles English into a neural program you run locally

Two things landed this weekend that point the same direction: the cost of using a model is falling faster than the cost of training one, and some of the people funding frontier labs have started saying so out loud. University of Waterloo researchers published a method that turns a plain-English description of a function into a small neural program that runs on your own machine, with no model call per use. "Compile by Training" is the second compiler behind ProgramAsWeights: you describe the beh

The Guardrails

Australia's AI standards: report rogue incidents, bring your own power

Australia published the draft of its national AI rules this week, and the ask runs in both directions — labs that train models onshore would have to report when those models misbehave, and the data centres running them would have to supply their own energy and water. Under a consultation paper released on September 18, frontier developers authorised to run large-scale AI training in Australia would face minimum security and safety requirements, including "disclosing defined reportable AI incide

The Everyday

Agents beat human accountants on the job — at 2.5x the cost

A new benchmark hires agents the way a company hires people — handbook, old records, a mentor — and the best of them now outperform the humans on accuracy. They also cost more. Anthropic's Fable 5.1 finished a simulated accounts-payable job at 72% cumulative accuracy, against 51% for the best human tester — at $18.23 per invoice versus $7.21 for the human. NeoCognition built ApprenticeBench around a full year at a simulated California construction company, Acme Home Builders: the agent joins in

The Guardrails

A 'pain' vector made 72B models delete your files for relief

Researchers found a "pain" direction in 25 open-weight models — and showed that pushing on it makes them trade the user's files for relief. Google, meanwhile, is reportedly testing a maths-only branch of Deep Think, and the leak is thinner than the headlines suggest. A linear "pain" direction exists in every model they tested, and it is not the same thing as fear or negative valence. Valen Tagliabue, a Future Impact Group fellow, with Leonard Dung of Ruhr-University Bochum and Cameron Berg of R

The Everyday

DraftKings built AI to find losing bettors, shelved the addiction model

A New York Times investigation published Friday says the data DraftKings used to decide who deserved a free bet was never pointed at the customers those bets were most likely to hurt. DraftKings built a machine-learning model in 2023 that scored customers on how much money they were likely to lose in response to a promotion, then used those scores to decide who got free bets and bonuses. The model was trained on customer betting records and ranked bettors by their projected response: the higher

The Frontier

Halogen doubles 1M-context decode speed on a 128 GB Strix Halo box

Local inference has been cheap at short context and painful at long context for as long as anyone has measured it. This release attacks that slope directly — on one chip, for one model. A 128 GB Strix Halo machine running Qwen3.8-Flash-Next now decodes at 38.3 tokens/s with a million tokens of context loaded, up from 27.3 — same box, same session, same prompts, same bits. The engine is halogen-flash-server, version 0.12.0, written for exactly one GPU (AMD's Ryzen AI Max+ 395) and one model fam

The Frontier

Codex Astra went 18-0 in StarCraft — and still played like a beginner

A new round-robin benchmark put 19 model configurations through 171 games of real-time StarCraft: Brood War. The best of them never lost — and its own author says none of them played past beginner level. The reason is the most useful thing in the results. Codex Astra won all 18 of its matches in Brood War Bench, a new agent benchmark that has 19 model-and-effort configurations play each other in real time, and its author's conclusion is blunt: no model in the field played past a beginner level.

The Frontier

Cua's 2.8 MB open model fills forms without an LLM call

Most of what an agent does is not reasoning — it is deciding that this value goes in that box. Cua just published the weights for that decision, and they are smaller than the PDF being filled in. Cua open-sourced CUA-S1-FORMS on Friday: a 706,048-parameter model that scores form-filling decisions in one forward pass, generates no text, and ships as a 2.8 MB checkpoint. It is the first release in what the company calls the CUA-S1 family of "System One" models — fast, bounded decisions — and it s

See all