Search
6 stories for“terminal-bench 4.0”
A cheaper, faster model arrives in ChatGPT Work, Codex and the API. OpenAI's own tests show big gains, but the error figures are internal and the company still sends hard coding elsewhere.
Oct 27, 2026
Released 3 September, Astra found genuine security flaws in testing and posted OpenAI's best-ever reasoning scores, but Artificial Analysis still ranks Claude Fable 5.1 ahead of it on broad intelligence and coding. Pricing matches Fable at 10 and 50 dollars.
Sep 24, 2026
Reflection AI released Beam under Apache 2.0 while Meta shipped Muse Spark Open under a tighter licence, giving developers two Western rivals to China's open models with different trade-offs.
Oct 18, 2026
Fable 5.1 and Mythos 5.1 are the same model with different safeguards. One is generally available. The other is restricted to a set of American organisations through trusted access programmes, for work in cybersecurity and the life sciences.
Sep 1, 2026
Richard Socher's lab says its automated research system did most of the work on a 40-billion-parameter model that activates 2.5 billion at a time and ships under Apache 2.0.
Oct 22, 2026
Crude shipments through the Gulf have recovered sharply, yet refineries and fuel supply remain constrained. That gap explains why cheaper oil may not quickly reach consumers.
Jul 31, 2026
