OpenCode's Mystery Omen Alpha Model Has Developers Guessing Who Built It

OpenCode released a stealth AI coding model called Omen Alpha on September 4, available exclusively to OpenCode Go subscribers for $10 per month with $100 worth of usage, but its identity may be revealed by a URL on OpenCode's own model-data page that files it under a 'zhipu' namespace, suggesting it was built by Zhipu AI, the company behind the Z.ai brand. This follows Zhipu's earlier reveal that its anonymous 'Ox Alpha' model on OpenRouter was actually GLM-5.3-Flash, an MIT-licensed open-weight model with 320 billion parameters (18 billion active), priced at $0.15 per million input tokens and $0.50 per million output tokens, which scored roughly 80% pass rate on DeepSWE testing versus Claude's 65%.

OpenCode just dropped a stealth AI coding model called Omen Alpha with no vendor name attached, and its own backend may have already given away who built it. On September 4, OpenCode announced Omen Alpha on X: a new stealth model, exclusive to OpenCode Go subscribers, offering $100 worth of usage for the $10 monthly plan. No benchmark chart. No blog post explaining the architecture. Just a name and a price. It's the same playbook OpenRouter and OpenCode have run all summer: let developers test frontier-caliber models blind, before a lab claims them. You can't find Omen Alpha on OpenRouter yet, unlike its predecessor. It lives inside OpenCode's own product, gated behind the Go subscription. So the fastest way to try it today is to sign up for that $10 tier and start pointing it at real code. The Zhipu Clue The identity clue didn't come from OpenCode's marketing. It came from the platform's own plumbing. OpenCode's internal model-data page for Omen Alpha sits at a URL that files the model under a 'zhipu' namespace: the same company, Zhipu AI, that operates under the brand Z.ai. Developers on the Chinese forum LINUX DO spotted it within hours, with one thread titled roughly 'anonymous model omen-alpha basically confirmed as a small Zhipu model.' Another asked plainly whether OpenCode's own webpage was showing it as Zhipu. That's forensic sleuthing built on a URL string, not a confirmation from Zhipu itself. Nobody at the company has said a word publicly. China's GLM-5.3-Flash Just Beat Claude Opus 4.8 on Agent Benchmarks https://startupfortune.com/chinas-glm-53-flash-just-beat-claude-opus-48-on-agent-benchmarks/ Zhipu AI revealed that its mysterious "Ox Alpha" test model was actually GLM-5.3-Flash, an MIT-licensed open-weight model that beats Claude Opus 4.8 on key agent benchmarks while costing roughly 42 times less. The release piles more pricing pressure on US labs just as agentic coding tools become the biggest driver of token spend. - GLM-5.3-Flash beats Claude Opus on agent benchmarks https://startupfortune.com/chinas-glm-53-flash-just-beat-claude-opus-48-on-agent-benchmarks/ - China's GLM-5.3-Flash model pricing and performance comparison https://startupfortune.com/chinas-glm-53-flash-just-beat-claude-opus-48-on-agent-benchmarks/ Z.ai's Playbook, Twice in a Month This isn't the first time this has happened. Two weeks earlier, on August 20, an anonymous model called Ox Alpha showed up on OpenRouter, free to use, with a 1 million token context window and no data retention. Developers ran it hard for six days before Z.ai confirmed to Bloomberg, according to reporting picked up by the South China Morning Post, that Ox Alpha was in fact GLM-5.3-Flash: a 320 billion parameter model with 18 billion active parameters, released under an MIT license and priced afterward at $0.15 per million input tokens and $0.50 per million output tokens. That reveal is exactly why the Zhipu guess for Omen Alpha carries weight. Z.ai has now run the same play twice in a month: stealth-drop a model on a coding platform, let the community stress-test it, then own up once the buzz builds. If Omen Alpha follows the same arc, an official confirmation could land within days. Early numbers explain why labs bother with this approach at all. In community DeepSWE testing after the Ox Alpha reveal, GLM-5.3-Flash scored roughly an 80% pass rate on a first attempt, ahead of Claude's 65%, GLM-5.3 and Grok 4.6 at 62% apiece, and GPT-5.6 at 52%, according to informal test threads developers shared after running the comparison themselves. Those figures come from small, self-reported samples, not a peer-reviewed suite. Treat them as directional. Still, a free model beating Claude on any coding measure is the kind of result that gets screenshotted and shared. That's the whole point of shipping it without a name attached. Omen Alpha hasn't produced numbers like that yet. Developers are running it through the informal 'Pelican test,' where you ask a model to draw a pelican riding a bicycle in SVG code and judge the output by eye, and at least one livestreamed DeepSWE test thread was underway on LINUX DO this week. No hard scores had surfaced by the time this article published. Why It's Worth Watching For founders and engineering leads evaluating coding assistants, stealth releases earn their keep for a simple reason. It's free signal. You get to see how a frontier-caliber model performs on your own codebase - no sales call, no annual contract - before a vendor sets a price that assumes you're already locked in. Ox Alpha proved that signal was worth having. It went from anonymous curiosity to a model people cite by name, GLM-5.3-Flash, in under a week. Whether Omen Alpha follows the same path depends on Zhipu, or whoever actually built it, deciding to step forward. Until then, the only confirmed facts are the ones OpenCode put in writing: a new model, a Go subscription, and $100 of usage for ten dollars. Also read: Dell's AI server backlog nearly doubled to $95 billion in one quarter https://startupfortune.com/dells-ai-server-backlog-nearly-doubled-to-95-billion-in-one-quarter/ • Anthropic Nears a $15 Billion Credit Line as Its IPO Push Accelerates https://startupfortune.com/anthropic-nears-a-15-billion-credit-line-as-its-ipo-push-accelerates/ • China's GLM-5.3-Flash Just Beat Claude Opus 4.8 on Agent Benchmarks https://startupfortune.com/chinas-glm-53-flash-just-beat-claude-opus-48-on-agent-benchmarks/ Zhipu's GLM-5.3-Flash Undercuts Claude and GPT on Price, Not on Hardware https://startupfortune.com/zhipus-glm-53-flash-undercuts-claude-and-gpt-on-price-not-on-hardware/ Zhipu AI revealed that its viral anonymous model Ox Alpha was actually GLM-5.3-Flash, an open-weight, MIT-licensed system that undercuts Claude and GPT on API price while matching them on coding benchmarks. Running it locally, though, still requires a serious multi-GPU setup, not a single consumer card. - GLM-5.3-Flash model pricing compared to Claude Opus https://startupfortune.com/zhipus-glm-53-flash-undercuts-claude-and-gpt-on-price-not-on-hardware/ - open weight model performance on coding benchmarks https://startupfortune.com/zhipus-glm-53-flash-undercuts-claude-and-gpt-on-price-not-on-hardware/ Founder discussion Open in the community → /community/ Almost there. Sign in and your reply posts straight away.