Skip to main content

The news desk

AI news, checked against the source.

What changed in AI, one story at a time. Each one links to its primary source, and facts stay apart from interpretation.

Last updated

Monday, October 5UTC

  1. New
    Reflection AIBeamModels

    Reflection previews Beam, a 501B coding model, with weights promised later

    Reflection announced Beam, a sparse 501B-parameter model built for coding and agent work. The company says early access is by signup while red teaming continues; weights, a model card and technical artifacts are promised later in October under Apache 2.0. This is a preview announcement, not a released open-weight model.

    Reflection AI1 source

Free, email only

Unlock the full brief free.

  • Thirty days of stories to browse, not seven
  • On every story: what to check, each source and why it matters
  • The same signup unlocks every free tool

This is not an account: there is no password, and the unlock is a cookie in this browser. You also join the Rise Productive newsletter from Demetri Panici, about once a week: what I built and what changed in AI. We'll email you a link to confirm, and you can unsubscribe in one click. The same signup unlocks every free tool on the site. How your email is handled.

Already subscribed? Enter the same email to unlock this browser. You won't be signed up twice.

Saturday, October 3UTC

  1. New
    Aleph AlphaKolibri 1Models

    Aleph Alpha releases Kolibri, open weights for German-English business AI

    Kolibri 1 is a German-English reasoning and tool-use model released with Apache 2.0 weights. Aleph Alpha positions it for sovereign and industrial deployments. Its 78.1-billion-parameter mixture-of-experts architecture activates about 3.46 billion parameters per token, but serving it still requires substantial GPU infrastructure; this is not a small laptop model.

    Aleph Alpha2 sources

Friday, October 2UTC

  1. New
    OpenAIChatGPT FinancesModels

    ChatGPT Finances expands to Free and Go users in the US

    OpenAI is rolling out Finances to Free and Go accounts in the United States on web, iOS, and Android. Connected financial information can ground explanations of spending, savings, and investments. This is a material plan-access expansion, separate from GPT-6 model launches, and not a money-transfer or trading agent.

    OpenAI2 sources

Thursday, October 1UTC

  1. New
    GoogleGuided Vision in Gemini LiveModels

    Gemini Live adds conversational visual assistance on Android

    Guided Vision gives spoken descriptions and camera-reframing cues during a Gemini Live session, with settings, accessibility shortcuts, and TalkBack entry points. Google says it is available on Android 9 and newer where Gemini Live is supported. It can make mistakes and is not a mobility aid, navigation system, or obstacle detector.

    Google1 source

  2. New
    CloudflareClef and Clef-flashModels

    Cloudflare releases typed decision models for agent routing and guardrails

    Cloudflare launched Clef and Clef-flash on Workers AI. They score allowed answers to typed questions rather than generating free-form prose, giving applications structured probabilities for routing, classification and action checks. The models support visual inputs and a 64k context window; official model pages and Cloudflare-owned Hugging Face distribution corroborate availability. Flash weights are published with an Apache-2.0 license. Fine-tuning starts as a hands-on service; a self-serve RL platform is future work, not an available turnkey product.

    Cloudflare4 sources

Wednesday, September 30UTC

  1. New
    GoogleGemini 4 ArgonModels

    Gemini 4 Argon targets long agent workflows, with access limited to trusted defenders

    Google announced Gemini 4 Argon for extended coding, enterprise work and defensive security, initially through its Fairwind trusted-defender program. Google describes a one-million-token output limit and reports internal agent-assisted code migrations. Wider developer and consumer access is planned, not generally available at announcement.

    Google3 sources

  2. New
    DigitalOceanGPT-6.1 Sol on DigitalOceanModels

    DigitalOcean adds GPT-6.1 Sol to its serverless inference catalog

    The dated provider notice makes GPT-6.1 Sol available through DigitalOcean Inference. This is a distribution and access change, not a new GPT-6.1 Sol model launch. Current usage and endpoint constraints must be checked in the supported-models reference; deprecated ADK deployment paths are not a recommended way to start.

    DigitalOcean2 sources

  3. New
    CohereEmbed 5 Pro / FastModels

    Cohere launches Embed 5 Pro and Fast for searching business documents

    Cohere released two embedding models for retrieval across text, images and multilingual business documents. Pro and Fast share an embedding space, allowing teams to index with Pro and query with either tier without rebuilding the index. The company lists availability through its API and Model Vault, Microsoft Foundry and Amazon SageMaker; its performance comparisons are vendor-reported.

    Cohere1 source

Days are UTC. Some sources report an exact time and others only a date; both are placed on their UTC day.