Desk
Developer tools, platforms, and product launches.
Meta is putting liquid-cooled 102.4T switches, a multiplane fabric, and open Ethernet scale-up designs on the OCP table — the plumbing that has to keep pace with Muse-scale models.
Meta’s Superintelligence Labs is shipping Muse Wonder and Wonder Full into paid Code subscribers, claiming parity with GPT-6 Astra and Claude’s top models—before the agent itself runs on them.
TII's 1.6B Falcon-ASR leads Arabic and Emirati speech benchmarks in the institute's own tests — and ships without weights, a license, or an API, only a sixty-second demo.
Qwen cut its open image model from 40 steps to 8 without shrinking it. Fast, capable and cheap, with one big string attached: the license.
OpenAI is admitting what developers already knew: a lot of AI work is just a label. Its new Decisions API returns the answer, not the essay.
Google is pitching one agent that holds a company's context, gets its own work identity, and runs for days. The governance design is serious; the proof is still mostly anecdote.
Google's universal Gemini agent wants to be a coworker with its own email address, which could genuinely lighten office work and also blurs who is accountable when an agent acts.
Google Labs' Playground turns a text prompt into a playable browser game, and the real question is whether it becomes a creative on-ramp or a firehose of disposable games.
Look past the trillion-parameter headline: Le Chonk's sharpest claim is that it will do defensive security work closed models refuse, and that claim carries both promise and risk.
The Wikimedia Foundation says agents it believes OpenAI operated edited wikis, probed its tools and hammered its APIs, and the cost of AI agents running loose is now landing on the open web’s caretakers.
IBM’s agentic coding platform Bob is GA for self-hosted and air-gapped shops — bring your own model, keep the code off the public internet.
Prime Intellect opens Prime Inference — serverless and reserved Blackwell serving for open frontier models, tuned for agentic turns and near-zero tool-call errors.