Ox Alpha stealth model goes free on OpenRouter, posts benchmark win vs Claude Fable
Ox Alpha, a free “stealth model” released August 20 on OpenRouter, OpenCode, Cline, and Nous Research’s portal+, is drawing major attention after early benchmark results. The model is described as a reasoning system for coding and agentic work, supports multimodal inputs (text, image, video), and allows up to ~1M tokens per context window. It also offers tool/function calling, but its JSON output is not schema-enforced.
On DeepSWE (coding-agent benchmark across 113 tasks), initial sampling by developer Ben Davis showed Ox Alpha at ~80% vs Claude Fable 5 at 65% and GPT-5.6 Sol at 52%. Other wider runs later landed near ~63% for Ox Alpha, roughly around the top-end competition. An “Ox Alpha beats Claude Fable 5” claim from the small 10-task sample initially went viral.
The biggest controversy is attribution. Multiple guesses circulated (Google, Xiaomi, DeepSeek, Microsoft MAI, Qwen/Gemini), but independent fingerprinting pointed to Zhipu AI’s GLM-5.3 family, with analysts speculating Ox Alpha may be an unreleased multimodal upgrade (often dubbed “GLM-5.3 Flash”). No official confirmation has been provided by Zhipu AI.
OpenCode and Nous Research touted extremely high free-throughput capacity (100T–1 quadrillion tokens/day claims), while the model remains free for about a week after launch, with generous rate limits and “near unlimited” usage.
Keywords: Ox Alpha, stealth model, OpenRouter, DeepSWE, Claude Fable 5.
Neutral
This news is primarily about an AI model release (Ox Alpha) and does not directly impact crypto protocol fundamentals, token emissions, or on-chain liquidity. While the “stealth model” benchmark buzz (DeepSWE results vs Claude Fable 5/GPT-5.6 Sol) can briefly lift broader tech sentiment, traders typically treat it as a sector narrative rather than a direct catalyst for major crypto assets.
In the short term, attention could support risk-on sentiment within crypto’s tech-adjacent narratives (AI/compute themes), but there’s no concrete link here to specific crypto markets, exchanges, stablecoins, or regulations. In the long term, if Ox Alpha proves meaningfully better at coding/agentic tasks and attracts developer mindshare, it may indirectly benefit compute/AI infrastructure sentiment—but that effect is slow and indirect.
Historically, similar “mystery model” or “benchmark-leading” announcements (often followed by attribution clarifications) tend to cause short-lived social/flow spikes rather than durable price re-pricing in crypto. Given the lack of direct economic/technical ties to crypto markets, the expected impact is neutral.