The Token Reporter

A day in AI

Wednesday, 21 October 2026

40 stories · 9 from multiple sources

▲ 2 signals · 2 sources

OpenAI models are coming to Google Cloud

OpenAI and Google Cloud are bringing frontier models and Codex workflows to Gemini’s Model Garden, available through Google Cloud Marketplace.

openai.com · 25 days ago

▲ 2 signals · 2 sources

Anthropic takes on Muse with Claude Duet, its new AI agent

theverge.com · 25 days ago

▲ 2 signals · 2 sources

Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster

huggingface.co · 25 days ago

More stories

▲ 2 signals · 2 sources

Walmart is inviting AI agents to shop for you

theverge.com · 25 days ago

▲ 2 signals · 2 sources

While Amazon blocks AI agents, Walmart rolls out the welcome mat

techcrunch.com · 25 days ago

▲ 2 signals · 2 sources

Patch the Planet: global response coordination as agents uncover critical vulnerabilities

openai.com · 25 days ago

▲ 2 signals · 2 sources

Using thousands of AI agents to uncover new leads in biological data

openai.com · 25 days ago

Tried running 748B DeepSeek-V4.1-Flash on a €2k PC

reddit.com · 25 days ago

Broad access through consistent standards

openai.com · 25 days ago

Also today

Physicist Christopher Stubbs joins Anthropic's Long-Term Benefit Trust

How AT&T is rethinking in-house legal work with OpenAI

Democratic governance of frontier AI, part II: Empowering cyber defenders

How Claude designed protein binders against scorpion neurotoxins

Today's PHP Security Release, and How We're Strengthening PHP Security

simonwillison.net

Panache — RTS Belle Époque pour mobile

simonwillison.net

Claude Duet

sparqlite: SPARQL 1.1 inside SQLite, written in Rust

simonwillison.net

DeepSeek just open-sourced its inference engine

Google staff resign in backlash over its AI contracts with military

Claude Duet

Canary Tokens

canarytokens.org

Anthropic's Public Benefit Charter

NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required

Train in Simulation, Deploy on a Real Robot: Microduck with Isaac Lab 3.0

How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications

Empowering frontline maternal care with AI-assisted ultrasound

research.google

Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.

Adaptive agentic intelligence framework using Strands Agents

Introducing Rerank 5: precise, efficient enterprise document retrieval

Train Qwen3 on AWS Trainium with torchtitan

Geezuz kernel: Qwen3.8-27B at 190 tok/s (620~ tok/s at 8 requests) on two 3090s

How U.S. Hunger connects leaders and data using Amazon Quick

Amazing free 3090 speedup hack, verified works 15% PP, 5% TG for free

Claude Mods: Flipping the Harness Inside Out

The revised rules of engineering management with Will Larson

newsletter.pragmaticengineer.com

When will we see these changes? --> China Is Doubling RAM Production to Keep Everyday Gadget Prices from Skyrocketing

Now that 3090 frenzy warning post makes sense.

OK, fine, Qwen 3.8 Flash Next is nice

RecursiveSI/flint (40B A2.5B) · Hugging Face

inclusionAI/Ming-3.0-TTS-0.5B

← All editions