Mystery model Ox Alpha tops GPT-5.6 — sleuths say it's Zhipu's GLM

Share
Mystery model Ox Alpha tops GPT-5.6 — sleuths say it's Zhipu's GLM

A nameless model dropped onto OpenRouter and immediately out-scored the biggest names in coding — and the detective work on its identity is as interesting as the model itself.

Ox Alpha, an anonymous "stealth model" that appeared on OpenRouter on August 20, is posting coding-test scores above GPT-5.6 Sol, and community forensics point hard at a Chinese flagship nobody has officially announced yet. OpenRouter lists it as a reasoning model built for coding, sustained agentic work and production workloads, with a 1-million-token context window (1,048,576 to be exact), 131K-token output, text plus native image and video input, and a price tag of zero during the preview — meaning the open-source coding agent OpenCode plugged it in the same day and provisioned roughly 100 trillion tokens a day of free capacity behind it.

Within hours the field was running it ragged. Independent researcher Ben Davis put it through 10 subtasks of the DeepSWE software-engineering benchmark and got an ~80% pass rate — well clear of Claude Fable 5's 65%, GLM-5.3 and Grok 4.6 at 62%, and GPT-5.6 Sol at 52%. None of that is official yet; Davis's run isn't on the DeepSWE leaderboard, and OpenRouter won't say who built the thing. So the community turned to fingerprinting: matching its tokenizer against known models (identical raw token counts to GLM-5.3, plus a fixed offset), comparing vision-encoder token budgets on scripted test videos (a perfect match with Zhipu's GLM-5V-Turbo on three independent design choices), and checking "dirty token" reactions, emoji density, and its ~117 agent steps per task versus GLM-5.3's 124. Davis's verdict sits at roughly 99% that Ox Alpha is Zhipu's GLM-5.x family — a multimodal GLM-5.3V or the next-gen GLM-5.5 — though Zhipu, Xiaomi, Tencent and even Google were all floated as suspects first.

The bigger pattern here is that stealth launches are quietly becoming the Chinese labs' go-to way into the overseas developer market. Ox Alpha is the fifth anonymous model OpenRouter has carried, and every prior one turned out Chinese: Pony Alpha became Zhipu's GLM-5, Hunter Alpha was Xiaomi's MiMo-V2-Pro, and Owl Alpha was Meituan's LongCat. An anonymous debut gets real-world feedback at scale without the brand baggage, then converts a mystery into a two-wave launch — first the model, then the reveal. With OpenRouter itself reportedly nearing a Stripe acquisition worth over $7 billion, the gateway to this traffic is also changing hands. The free preview window likely closes around August 27; keep an eye out for whichever lab steps out from behind the curtain before then.

Who do you think is behind Ox Alpha — Zhipu, or someone else entirely? Tell us in the comments.

Read more

Underdog launches a private on-device AI assistant, backed by a16z

Underdog launches a private on-device AI assistant, backed by a16z

The privacy split in consumer AI got a new entrant tonight, GitHub turned code-review benchmarks into a vendor-bias argument, and OpenAI opened its oddest API to everyone. Underdog launched in invite-only beta: an on-device AI assistant that keeps your data on your machine and never charges you a subscription. Self-taught coder and Thiel Fellow Sigil Wen — who moved to Silicon Valley at 17 and lived in an AI hacker house with Andrej Karpathy — built his own inference engine, Husky, to run a 2

Today in AI — October 6, 2026

Today in AI — October 6, 2026

A day of second-order moves: the labs are buying task data from software vendors instead of scraping the web, Waymo is borrowing to fund the robot world, and the biggest bank in the US just put a number on what Anthropic's latest model cost it in risk. Models & Research * OpenAI is training GPT-6 Astra on Ironclad's real contracting work. Ironclad staff helped turn 11 tasks across legal, commercial and procurement work — setting up NDAs, approval workflows, clauses that change by jurisdict

Meta, Walmart and Stripe publish the Personal Agent Protocol

Meta, Walmart and Stripe publish the Personal Agent Protocol

The agent economy is writing its rulebook tonight: one open standard for AI bots at the checkout, a cheaper image model from Google, and Anthropic turning bug-hunting into a tiered product. Meta, Walmart, Stripe and Sierra are publishing an open "personal agent protocol" — a standard that defines how personal AI agents interact with businesses online. The group behind it reads like a cross-section of agentic commerce: Meta, Sierra, Walmart, Stripe, Shopify, Genesys, Rocket, NiCE, Decagon and I

Lambda raises up to $4B from Blackstone ahead of its IPO

Lambda raises up to $4B from Blackstone ahead of its IPO

The neocloud money is consolidating fast, and today's inbox shows both ends of the market: a heavyweight pre-IPO round on one side, and a Google open model you can run on a phone on the other. Lambda is raising up to $4 billion led by Blackstone and Coatue at a $14.5 billion pre-money valuation — its last private round before a planned IPO. The Wall Street Journal reported the scoop from a letter to limited partners, and Reuters independently confirmed the headline terms: the round is led by t