Morning Edition · Saturday, July 4, 2026
Tech
Anthropic Ships Claude Sonnet 5, Pushing Near-Flagship Agent Performance Down the Price Curve
The new mid-tier model posts coding and agent scores that approach Anthropic's Opus flagship while launching at a fraction of its cost.
Tech
OpenAI Moves GPT-5.6 From Government-Gated Preview Toward Broad Release
After shipping its new Sol, Terra, and Luna models to about 20 vetted partners under a US safety review, OpenAI is preparing wider access with more generous usage limits.
Tech
Anthropic Redeploys Fable 5 and Proposes a Cross-Lab Rubric for Scoring Jailbreaks
With Amazon, Microsoft, and Google, Anthropic put forward a four-criteria severity scale for AI jailbreaks as it restored a model pulled under export controls.
Tech
Mistral Open-Sources Leanstral 1.5, Saturating a Formal-Math Benchmark
The Apache-licensed Lean 4 proof model solved 587 of 672 PutnamBench problems and reports finding five previously unknown bugs across real code repositories.
Tech
Meta's Brain2Qwerty v2 Decodes Typed Sentences From Brain Scans at 61% Word Accuracy
A non-invasive magnetoencephalography pipeline, with code and data released, narrows the gap to surgical implants while remaining pure research.
Markets
Thinking Machines and Bridgewater Say a Small Custom Model Beats Frontier Models on Finance Tasks
A fine-tuned model built on Bridgewater's expert-labeled data reportedly cut errors by nearly 30% versus GPT, Claude, and Gemini at a fraction of the cost.
Tech
Anthropic Launches Claude Science, Betting on Workflow Over a New Model
The workbench wires generalist agents into genomics, proteomics, and cheminformatics tools, with a reviewer agent that checks citations and calculations.
Tech
Google Invests $75 Million in A24 for a DeepMind Filmmaking Partnership
The deal gives DeepMind access to A24's production process to build generative tools, and has drawn immediate pushback from some of the studio's filmmakers.
Tech
Google Research Ships TabFM, a Zero-Shot Foundation Model for Tables
Trained entirely on synthetic data, TabFM predicts on unseen tables in a single forward pass and reportedly beats gradient-boosted trees on dozens of datasets.
Geopolitics
Thiel Attacks Pope Leo's Call for AI Rules, Casting Regulation as a Gift to Beijing
At the Aspen Ideas Festival, the investor said the pope's demand for international AI oversight would only slow the American side of the race with China.
Tech
Sakana AI Presents Sheaf-ADMM Method for Multi-Agent Coordination at ICML
The paper treats intelligence as a distributed system, splitting a hard problem into overlapping fragments that agents solve and reconcile.
Tech
Altman Promises Systems of 'Astonishing Power' as Scrutiny of AI Narratives Sharpens
The OpenAI chief's latest forecast arrives amid a market increasingly willing to separate verifiable capability from the rhetoric used to raise capital.