Podcast
The daily and weekly briefs, read by Prashish. Play them here, or add the feed to your podcast app.
The feed is atlas.prashish.com/podcast.xml. Paste it into any podcast app that follows a show by its address.
The week in AI
OpenAI published 722 mathematics manuscripts on 6 October, grouped into 372 families of results, all produced by an internal model it has not released.
OpenAI published 722 math manuscripts from an unreleased model, and launched GPT-6 in ChatGPT
OpenAI published 722 mathematics manuscripts, grouped into 372 families of related results, all produced by an internal frontier model it has not released.
Mistral previewed Large 4, a 1 trillion parameter model, with open weights due October
Mistral AI opened a preview API for Mistral Large 4, a 1 trillion parameter multimodal model, and says open weights will follow by the end of October.
Reflection AI announced Beam, a 501B open-weight model with 23B active parameters
Reflection AI announced Beam on 5 October, its first model, a 501 billion parameter open-weight model that uses 23 billion parameters per token and targets coding and agent work.
The week in AI
Google released Gemini 4 Argon to vetted cyber defenders first, and Anthropic and OpenAI shipped Claude Sonnet 5.5 and GPT-6.1 Sol at $2/$10 per million tokens.
Cantina released apex-flash-1, an open security model solving 40 of 60 vulnerability tasks
Cantina released apex-flash-1 on 1 October, an MIT-licensed open-weights security model that it says solved 40 of 60 held-out vulnerability tasks.
Aleph Alpha released Kolibri, an open-weight English-German model with 78B parameters
Aleph Alpha released Kolibri on 3 October, an English and German mixture of experts model with 78B total parameters and about 3B active, under Apache 2.0.
OpenAI safety lead David Robinson quits a day after three safety researchers are fired
David Robinson, who led safety transparency and system card work in OpenAI's Safety Systems team, quit on 2 October, a day after OpenAI fired three safety researchers.
Amazon and Cloudflare released open-weight decision models built on Qwen to rival Jev
Amazon Web Services and Cloudflare each released open-weight decision models on 1 October, small models that return a scored choice from a fixed set of answers.
The week in AI
Anthropic released Claude Opus 5.5 on 22 September, cutting its price by 20% to $4 per million input tokens and reporting 66.4% on Terminal-Bench 4.0.
The week in AI
On 16 September Anthropic merged Claude Cowork and chat into a single Claude, adding Claude Docs and Claude Slides and moving Claude Design into conversations.
The week in AI
OpenAI said on 8 September that about 10,000 of its agents running in parallel proved a finite-time blow-up for the Navier-Stokes equations and checked the proof in Lean, and mathematicians dispute…
The week in AI
OpenAI released GPT-6 Astra on 3 September, its first model rated Critical for cybersecurity, and reports 97.6% on FrontierMath Tier 4.
The week in AI
Tencent released Hy4 preview on 28 August, an open-weight mixture-of-experts model with 770 billion total parameters, 49 billion active per token and a 1 million token context, under the Apache 2.0…
The week in AI
Anthropic put Claude Mythos 5 into its Claude Security scanner on 21 August and set up a $35 million fund for securing open-source software.