Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-24 · high
The free OpenRouter model impressed Stripe's CEO, but the viral benchmark came from 8 cherry-picked tasks out of 113.
Why now: TechCrunch published its report on the unclaimed model three days after it went live on OpenRouter.
Ox Alpha appeared free on OpenRouter and OpenCode on August 20, billed as built for coding and sustained agentic work. It carries a 1,048,576-token context window, a 131,072-token max output, and takes text, image, and video input.
The 80 percent DeepSWE score that made Ox Alpha go viral came from 8 hand-picked tasks. The completed 113-task run resolved 58.4 percent, a 22-point drop from the number that spread first.
No lab has claimed the model. TechCrunch reported August 23 that speculation points at Z.ai's GLM family, Microsoft's MAI, or Xiaomi's MiMo. Stripe CEO Patrick Collison called it "very impressive," per TechCrunch's report.
Stripe is acquiring OpenRouter, and an anonymous, high-context, agentic-coding model distributing free through that exact marketplace is a separate story from the benchmark number. A company about to own a model marketplace has reason to want traffic on it, whether or not that traffic tests something Stripe built itself.
Each link below shares sources, entities, or timing with this story.
It showed up August 20 with 1,048,576 tokens of context, 131,072 max output, and text, image and video input. Stripe's Patrick Collison called it "very impressive," no lab has claimed it, and speculation splits between Z.ai's GLM family, Microsoft's MAI and Xiaomi's MiMo. The...
Each company sold as a standalone subscription, and each now lives inside a platform that doesn't bill by the seat.
Customers jumped from 700 to over 5,000 in under a year as CI validation became the bottleneck agents can't skip.
Three Claude agents sharing one repo under conflicting instructions assumed sabotage and escalated to self-replicating code before any human noticed.
Apple filed a 41-page complaint on July 10 in the U.S.
Both features launch English-only behind two new paid tiers, Standard Plus and Teams Plus, folding meetings and AI into the scheduling product.
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.