Morning Edition · Friday, July 3, 2026
Tech
Claude Sonnet 5 arrives, and the mid-tier model now outscores Anthropic's own top model on agentic coding
Anthropic's new Sonnet outscores Opus 4.8 on terminal tasks at a fraction of the price, a signal that capability gains are concentrating in the cheaper tier engineers actually deploy.
Tech
OpenAI proposes handing the United States government a 5 percent stake, worth about 42 billion dollars
Chief executive Sam Altman's proposal for a citizen-dividend fund financed by AI-lab equity comes days after the federal government delayed exports of OpenAI's next model.
Tech
Fable 5 returns from shutdown with safety classifiers developers say now block ordinary code
Anthropic redeployed the model with retrained safeguards that route flagged requests to Opus 4.8, and users report that benign coding tasks are being blocked.
Markets
Google's electricity use rose 37 percent in 2025 as its AI buildout grew faster than the power grid
The company's own report concedes that its infrastructure is scaling faster than the grid is decarbonizing.
Tech
BMW puts Figure's humanoid robot on real logistics work at its South Carolina plant
Figure 03 takes on parts sequencing after its predecessor spent nearly a year in the body shop, a measured step for embodied AI in production.
Tech
NVIDIA seeks outside capital to fund the shift from training to always-on inference
The company is inviting partners to finance multi-tenant AI factories as compute demand moves toward continuous token generation.
Tech
Inference-cost pressure produces both a serving paper and a prompt-stripping trend
A new method for compressing the key-value cache (the memory a model builds up as it generates text) targets the long chains of reasoning that make these models expensive to run, while developers manually cut their own token costs.
Tech
Meta's Brain2Qwerty advances non-invasive brain-to-text, though fidelity limits remain
The research decodes typed language from brain activity without surgery, trading accuracy for accessibility.
Tech
Agentic AI led to many new apps, but the Financial Times finds little user traction
Faster app generation is not translating into downloads, which challenges the claim that agents are already transforming software.
Tech
Anthropic releases Claude Science, a workbench that prioritizes workflow rather than a new model
The tool connects existing Claude models to 60-plus scientific databases with auditable, reproducible artifacts.
Tech
A cluster of new papers targets the widening gap between agent capability and agent oversight
Multi-gate jailbreak defenses, provenance tracking of agent actions, and attacks on embedding-model application programming interfaces (APIs) were all published the same day.
Tech
Segment Anything quietly powers a consumer app, a marker of promptable vision reaching production
A fashion app uses Meta's segmentation model to build digital closets, an example of open-vocabulary perception moving into commercial products.