Osworld Verified

Summary: Qwen3.8-Max in 30 Seconds Qwen3.8-Max is Alibaba’s new flagship model, made generally available on August 2, 2026. 2.4 trillion parameters, 95 billion active (MoE architecture). It reads text, images and video, and returns text. Context window is in the 1 million token class. On one task it ran 125 hours (about 5 days) with no human input, rebuilding an experiment from a machine learning paper from scratch, confirming its six findings, then inventing a method that beats the paper. It beats Claude Opus 4.8 on most coding and agent tests, trades blows with Claude Fable 5 and GPT-5.6 Sol, and falls behind on some. API pricing is $2 input / $6 output per million tokens. Repeated input costs $0.25. This is the first time Alibaba will open-weight a Max-class model. “Next week” is the word, but there is no date, license or model card yet. Ask an AI to “rebuild the experiment in this paper, then improve on it” and it normally stalls after a few turns, waiting for you to step in and steer. ...

On July 21, 2026, Google announced three new Gemini models at once: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and the defense-focused Gemini 3.5 Flash Cyber. The theme of the announcement is clear: not higher scores, but getting the same job done with fewer tokens. As agentic workflows move into production, the bill is driven less by how smart a model is and more by how many tokens that intelligence burns. Google aimed straight at that. Here are the details. ...

Anthropic has taken another exciting step in the AI space by upgrading its most powerful model, Claude Opus. Meet Claude Opus 4.8! Built on the foundations of Opus 4.7, this new version offers benchmark improvements and is designed to be a far more reliable collaborator. Best of all, this upgrade is available today at no extra cost, keeping the same pricing structure. Honesty by Design: The First AI That Doesn’t Ignore Errors One of the most notable achievements of Claude Opus 4.8 is its progress on AI hallucinations and overconfidence. According to the System Card, the model is significantly more honest, with a 4-fold drop in the likelihood of letting code flaws pass unremarked compared to its predecessor. ...