---
format: "aidr-story-markdown/v1"
id: "1c098538b8441cec927daf3f13d33e0f3c008d6b9207a96d6ac8591781165951"
canonical_url: "https://aidr.today/1c098538?lang=en"
title: "Ox Alpha Hits 80% on Coding Benchmark, Besting Fable in Stealth Launch"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-21T23:49:27.000Z"
category: "Models"
topics: ["coding","benchmark","agent","llm"]
source_urls: ["https://huggingnews.com/ai/ox-alpha-hits-80percent-on-coding-benchmark-besting-fable-in-stealth-lau-a18bf2ea","https://x.com/alexatallah/status/2090828403486208361","https://x.com/alexatallah/status/2090835555151925703","https://x.com/AskVenice/status/2090847078633156947","https://x.com/cline/status/2090854216399220985","https://x.com/NousResearch/status/2090899914700054780","https://x.com/MTSlive/status/2090872472174522701"]
summary: "An anonymous AI model called Ox Alpha is now available for free across OpenRouter, Venice, and Cline, offering multimodal capabilities and a 1 million token context window. The stealth release scored 80% on a DeepSWE coding subset, outperforming rivals Fable at 65%, GLM-5.3 at 62%, and Sol at 52%. The model handles text, image, and video inputs and is designed for production use and agentic work. The rollout marks the fifth anonymous model drop in six months, a pattern previously associated with Chinese laboratories including Zhipu, Xiaomi, Ant, and Meituan. Among the entities providing access is the Nous Portal, which claims it can handle 1 quadrillion tokens per day for the model."
---

# Ox Alpha Hits 80% on Coding Benchmark, Besting Fable in Stealth Launch

> [Open the canonical story](<https://aidr.today/1c098538?lang=en>)

**Published:** 2026-08-21T23:49:27.000Z
**Category:** Models
**Topics:** coding, benchmark, agent, llm

## Summary

An anonymous AI model called Ox Alpha is now available for free across OpenRouter, Venice, and Cline, offering multimodal capabilities and a 1 million token context window\. The stealth release scored 80% on a DeepSWE coding subset, outperforming rivals Fable at 65%, GLM\-5\.3 at 62%, and Sol at 52%\. The model handles text, image, and video inputs and is designed for production use and agentic work\. The rollout marks the fifth anonymous model drop in six months, a pattern previously associated with Chinese laboratories including Zhipu, Xiaomi, Ant, and Meituan\. Among the entities providing access is the Nous Portal, which claims it can handle 1 quadrillion tokens per day for the model\.

## Sources

- [Story source](<https://huggingnews.com/ai/ox-alpha-hits-80percent-on-coding-benchmark-besting-fable-in-stealth-lau-a18bf2ea>)
- [Story source](<https://x.com/alexatallah/status/2090828403486208361>)
- [Supporting source](<https://x.com/alexatallah/status/2090835555151925703>)
- [Supporting source](<https://x.com/AskVenice/status/2090847078633156947>)
- [Supporting source](<https://x.com/cline/status/2090854216399220985>)
- [Supporting source](<https://x.com/NousResearch/status/2090899914700054780>)
- [Supporting source](<https://x.com/MTSlive/status/2090872472174522701>)