Thoughts, stories and ideas._
Anthropic says Claude Haiku 5.5 is its first Haiku-class model with an adjustable effort setting. Thinking is now adaptive and on by default, and the old token-budget setting returns an error.
11 Oct 2026Reflection AI has announced Beam, a text-only open-weight model. The company says it scores comparably to GLM-5.2 on reasoning at lower inference compute. Its own tables show GLM-5.2 ahead on four reasoning tests.
10 Oct 2026Nikkei Asia reports that Apple has told some suppliers to cut component production for the iPhone 18 Pro and Pro Max after memory-driven price rises dampened demand. Reuters could not immediately verify it.
9 Oct 2026The Wikimedia Foundation says it found sandbox edits, probing of a note-taking tool and heavy traffic that it believes came from OpenAI-operated agents. It found no evidence of compromise. OpenAI did not immediately respond to Reuters' request for comment.
9 Oct 2026OpenAI is bringing GPT-6 to ChatGPT on free and paid plans, with a feature called Intelligent UI that answers with buttons, forms and charts. Here is who gets which model, and what the system card says.
8 Oct 2026Mistral has launched a public preview of Mistral Large 4, a 1-trillion-parameter model. Its benchmark claims are Mistral's own, and the open weights are still to come.
8 Oct 2026David Robinson, who led the writing of OpenAI's launch safety reports, resigned and wrote in The Atlantic that the problem is culture, not rules. OpenAI says it pauses or holds back models when it needs to.
7 Oct 2026Anthropic's expanded Cyber Verification Program gives vetted security professionals access to Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1 with reduced safeguards — in three tiers, with the top tier reviewed alongside the US government.
7 Oct 2026