Wednesday Aug 26
COOL TOOLSOther

Nobody Knows Who Made Ox Alpha

26AUG
FREE MODELYOUR CODE

A free coding model showed up on model marketplace OpenRouter with no company name. It beat the big labs on a 10-task sample. The full benchmark says otherwise.

Developer Ben Davis ran 10 tasks from DeepSWE, a software engineering test. Ox Alpha hit 80 percent. Claude Fable 5 got 65. On the full 113-task run it scored 58.4.

Then developers went hunting. Tokenizer fingerprints and video encoder behavior point at Zhipu AI in Beijing. No lab has confirmed anything.

The free window closes around August 27. Prompts sent through OpenRouter are retained by a provider that will not name itself. Read the terms before you paste your codebase in.

full brief & sources

⚡ Why this matters

  • A frontier-class coding model is being handed out free by an operator nobody can name.
  • The gap between the 10-task headline and the full-benchmark number is a lesson in how AI benchmarks get sold.
  • Anyone piping a work codebase into a free endpoint is making a data decision without a counterparty.

🔍 What happened

  • Ox Alpha appeared on OpenRouter and OpenCode on August 20, 2026, free for roughly a week.
  • It offers a 1,048,576-token context window and accepts text, images and video.
  • Ben Davis ran a hand-picked 10-task DeepSWE sample: Ox Alpha 80%, Claude Fable 5 65%, GPT-5.6 Sol 52%.
  • On the full DeepSWE run of 113 tasks across 91 repositories, Ox Alpha scored 58.4%.
  • The official DeepSWE leaderboard lists Claude Opus 5 at 74% and GPT-5.6 Sol at 73%.
  • Tokenizer and video-encoder fingerprints point at Zhipu AI's unreleased GLM flagship. Zhipu has not confirmed.
  • The OpenRouter route retains prompts and completions; the OpenCode route states zero retention. Two routes, two policies.

💬 Smart takes

  • Ben Davis: ran the 10-task sample that produced the 80% number now quoted everywhere.
  • The benchmark read: a hand-picked 10 of 113 tasks can favor any model through selection alone.
  • Skeptic: retention by an unnamed provider means you cannot read the policy, cannot request deletion, and cannot name who holds your code.

🧭 Where this goes

  1. Likelya named lab claims Ox Alpha within weeks, as happened with earlier stealth models.
  2. Likelythe free window closes and the model reappears on paid pricing under a real name.
  3. PossibleZhipu confirms and this becomes a GLM launch story rather than a mystery.
  4. Wild Carda company discovers proprietary code went through the OpenRouter route and it turns into a compliance incident.

🥄 The Spoon Take

Two separate things got fused into one headline. A stealth model scoring well on someone's 10-task sample is not the same as beating Opus 5. The full run puts Ox Alpha 15 points behind. What is actually interesting is the setup: free frontier-grade inference, anonymous operator, prompts retained. That trade is the news, not the leaderboard.

🤔 Pushback

Stealth previews are a normal pre-launch practice and OpenRouter's terms are public. A 58.4% full run is respectable for an unannounced model. The hype is the coverage, not the model.

·TechTimes - retains every prompt·Decrypt - nobody knows who built it·OpenRouter - stealth model terms·BigGo - Zhipu fingerprint