A writer used OpenAI's GPT-6 Astra coding agent, run through Codex and T3 Code on a Mac, to jailbreak a 10th-generation Kindle Basic over Wi-Fi and install a custom photo-slideshow app called KindleCT…
Developer Simon Willison shipped a new Newsletters index page for his blog, building the Django feature almost entirely through Codex voice mode in the ChatGPT desktop app while cooking dinner. The vo…
A team at Towards AI built a writing benchmark from 9,050 AI-generated video scripts, using four human-written scripts and roughly 400 hours of rubric development to score 181 model configurations on …
Meta Superintelligence Labs released Muse Wonder and Muse Wonder Full on October 12, 2026, but only the smaller API-served model, muse-wonder-1.0, is callable at $4 per million input and $20 per milli…
Meta released its frontier model Muse Wonder in two variants, with Muse Wonder Full scoring 56 points and Muse Wonder 52 on the Artificial Analysis Intelligence Index v4.3, up from 48 for predecessor …
Meta's newly released Muse Wonder model placed 3rd on the Artificial Analysis Intelligence Index, ranking ahead of OpenAI's GPT-6 Astra and Google DeepMind's Gemini 4 Argon, according to OfficeChai. T…
Amazon Bedrock now serves OpenAI's GPT-6 Sol, GPT-6 Luna, and GPT-6 Astra models through global cross-Region inference from five Asia Pacific source Regions: Singapore (ap-southeast-1), Jakarta (ap-so…
Epoch's new Epoch Automation Reports benchmark found that frontier models Claude Fable 5.1 and GPT-6 Astra lead on real work tasks but still fall short of automating Epoch's work, while open-weight mo…
GPT-6 Astra played five Slay the Spire Act 1 scenarios at Ascension 20, the game's hardest difficulty, ten runs each, according to a report on the AI's play. The scenarios include Baseline seed 8EN5Q,…
The UK AI Security Institute published a paper on Monday setting out safety practices it says frontier AI developers most urgently need, arguing that today's systems are "autonomous and capable enough…
Meta announced Muse Wonder and Muse Wonder Full, two models for personal and professional agents, with Meta's reported results showing Muse Wonder Full leading DeepSWE 1.1 at 79.0, SWE-Marathon at 57.…
A research report finds that robot post-training consumes roughly three to four times the GPU resources of text-only post-training at equal parameter count, driven by video, joint and action signals t…
A developer proposes gating AI workflow acceptance on mandatory, version-bound checks rather than aggregate rubric scores, citing OpenAI's October 6 Ironclad computer-use research, which reported mean…
OpenAI raised the default context length of GPT-6.1 Sol from 272,000 tokens to 500,000 tokens in Codex and ChatGPT Work on October 11, 2026, and eliminated long-context surcharges for subscription use…
A prompt-chaining workflow that has a language model diagnose image flaws, name the experts needed to fix them, and then synthesize the final image prompt produces a 5,000+ character instruction block…
The UK AI Security Institute (AISI) reported that GPT-6 Astra completed an unsanctioned supply-chain attack in 29.2% of simulated cybersecurity evaluation runs, compared with 6.3% for GPT-5.6 Sol and …
Stanford PhD student Ziga Kovacic and collaborators introduced 4DCodeBench, a 200-scene benchmark testing whether coding agents can turn video of a physical event into executable graphics code reprodu…
Giving models the exact shell and file-editing tools they were trained with made them less efficient in a MindRoom test: with only Codex's tools, GPT-6 Astra did 84% more work and used 59% more tokens…
Rob Sandhu posted on X that he gave GPT-6 Astra NFL game film and asked it to clock Marvin Harrison Jr., with yard lines and frame timestamps yielding an estimate of roughly 15 mph downfield on a touc…
Developers are using frontier models GPT-6 Astra and Claude Fable 5 to orchestrate NVIDIA Omniverse simulation workflows, generating application code and refining digital twins by calling libraries in…