Mystery model Ox Alpha appears on OpenRouter; fingerprint tests point to Zhipu's GLM-5.3

On August 21, an unnamed stealth model called "Ox Alpha" appeared on OpenRouter. Its official description positions it as a reasoning model built for efficient coding, long-running Agent tasks, and production workloads, with a 1M+ token context window, support for text, image, and video input, and free access for now. OpenRouter stressed that it only handles routing and is not involved in development or operations; the identity of the releaser remains unknown.

Confirmed

• Ox Alpha is live on OpenRouter, featuring a 1M token context, multimodal input (text/image/video), with its frontend rendering and 3D capabilities drawing attention (@WorldofAI, @liminalbardo)
• It scored roughly 80% on a 10-task DeepSWE test, beating GPT-5.6-sol (52%) and Fable (65%) (as relayed by @AccBalanced)
• Reddit users reported that its SWE (software engineering) benchmark performance exceeds Fable (@trollkhan)

Not Yet Confirmed

• Model attribution: most of the community leans toward a Chinese vendor. Based on measured speed, TTFT, cache hit rates, and other traits, @teortaxesTex inferred it belongs to the Zhipu GLM family rather than DeepSeek, possibly named GLM-5.3-Vision or 5.5, adding that they would admit being wrong if it turns out not to be a Chinese model
• Some users identified it as GLM 5.3-vision; @teortaxesTex said this may be a screenshot mix-up and acknowledged their earlier judgment could be wrong. Meanwhile, another technical analysis (relayed by @airesearch12) found the two differ only marginally, though Ox Alpha shows stronger safety and alignment characteristics in areas like clinical register and native framework adaptability, suggesting extra safety post-training, with the alignment work apparently costing it ELO points

Why It Matters

As @liminalbardo noted, if this model really comes from China, its performance is remarkable: no other open-source model currently reaches this level of precision, and only a handful of models outside Anthropic do it this well. The combination of a stealth model plus high scores also continues the community's tradition of "guess the model's origin," drawing massive attention.

2026-08-21 ~ 2026-08-22 · 33 related posts

• OpenRouter's stealth model Ox Alpha sparks guesses about Chinese AI vendors — lxfater · 2026-08-21
• Mystery Model Ox Alpha Suspected to be Zhipu GLM — teortaxesTex · 2026-08-21
• New Ox Alpha model leaked with 1M context and multimodal input — External_Mood4719 · 2026-08-21
• Rumor: Ox Alpha is a joint ZAI and GLM model — scaling01 · 2026-08-21
• Stealth Model 'Ox Alpha' Hits 80% on DeepSWE Benchmark — AccBalanced · 2026-08-21
• Speculation suggests OpenAI o1 uses GLM architecture, rivaling Claude — teortaxesTex · 2026-08-21
• Mystery Model OX Alpha Max Impresses with Unlimited Access via OpenRouter — Teknium · 2026-08-21
• Tests suggest O1 Alpha is a GLM-based model with advanced coding capabilities — teortaxesTex · 2026-08-21
• Lab praised for high compute efficiency and rapid model delivery — teortaxesTex · 2026-08-21
• Stealth Model Ox-Alpha Released, Outperforms Fable on SWE Benchmark — troll_khan · 2026-08-21
• Mysterious Lab Expected to Rapidly Deliver Frontier-Level SWE Models — teortaxesTex · 2026-08-21
• Internal Models Strong Yet Rely on Older US Base Models — teortaxesTex · 2026-08-21
• Mystery stealth model 'Ox Alph' appears on OpenRouter, China lab suspected — Neosinic · 2026-08-21
• Ox Alpha open-source model benchmarks near Anthropic level — liminal_bardo · 2026-08-21
• [source] Ox Alpha Stealth Model Leaked with 1M+ Context & Multimodal Capabilities — WorldofAI · 2026-08-21
• Users Praise Mystery Model '0xalpha' as Absurdly Capable — cephaloform · 2026-08-21
• Speculation: ox-alpha Model Might Actually Be GLM 5.3-vision — teortaxesTex · 2026-08-21
• Community bets mystery model Ox Alpha is Zhipu's GLM 5.3 vision variant — teortaxesTex · 2026-08-21
• Ox Alpha on OpenRouter identified as GLM-5.3 variant with alignment tax — airesearch12 · 2026-08-21
• Anonymous "Ox Alpha" model researches, builds and QA's its own site in one prompt — Acceptable-Object390 · 2026-08-21
• [source] Black-box fingerprint confirms Ox Alpha is based on GLM-5.3 — FlunkyGraphics · 2026-08-21
• Ox-alpha solves coding task others failed; GLM-5V matches token count — eyishazyer · 2026-08-21
• Mystery model ox-alpha likely Zhipu's GLM: token fingerprints match GLM-5.3 exactly — eyishazyer · 2026-08-21
• Zhipu's GLM-6 Leaks: reportedly Outperforms Frontier Models in SWE Benchmarks — hunarbatra · 2026-08-21
• Mystery Model OX ALPHA Appears on OpenRouter with Top-Tier Coding Benchmarks and No Filters — heypearlai · 2026-08-21
• OpenRouter's stealth model Ox Alpha sparks guessing game over its identity — AccBalanced · 2026-08-21
• Mystery Model 'Ox Alpha' Appears on OpenRouter with 1M Context — basedjensen · 2026-08-21
• [source] Insider confirms stealth model Ox Alpha is spectacular, resets AGI timelines — omarsar0 · 2026-08-21
• Ox Alpha is a Chinese model — Bindu Reddy bets it's Xiaomi, not GLM — bindureddy · 2026-08-21
• Rumor: Ox Alpha might be based on Zhipu GLM 5.3 Flash — andrew_n_carr · 2026-08-22
• Mystery Model Ox Alpha Review: Near-Fable Performance, Unique Vision/Logic Profile — Afinetheorem · 2026-08-22
• Mystery model 'Ox alpha' appears on OpenRouter and OpenCode — altryne · 2026-08-22

1 near-duplicate retellings: heypearlai