Anthropic launched Claude Opus 5 as a higher-performing Opus-tier model for coding, professional work and long-running agents. The company priced it at $5 per million input tokens and $25 per million output tokens and made it available across its platforms.
A software essay questions why rapid adoption of coding agents has not eliminated unreliable applications. It argues that greater output, expanding complexity and rising expectations can preserve or worsen quality problems even when models help engineers find bugs.
A personal essay describes a decline in sustained attention and experiments with timers, reduced phone access and deliberate limits on distractions. It presents an individual experience and practical reflection, not a clinical finding about cognition.
A report highlighted estimates of extensive off-balance-sheet obligations connected with data-center expansion at five large US technology companies. It cited an estimated $1.65 trillion outside reported balance sheets and compared that with $1.35 trillion of officially reported debt.
Author Neal Stephenson described his long-running practice of drafting books by fountain pen and offered practical advice on pens, paper, pressure and friction. He connected handwriting with coordinated physical and mental activity while clearly noting that he is not a brain expert.
On 2026-07-22, available reporting documented i generated 1,000+ SVGs across 7 frontier models to test whether AI labs are training on Simon Willison’s pelican-riding-a-bicycle benchmark.
A repeating wobble measured in a substellar companion points to a Jupiter-scale body circling it, but researchers say familiar Solar System labels may not fit.
An industry essay argues that portable Chinese models could gain distribution advantages over closed American services as performance differences narrow.
On 2026-07-20, available reporting documented everyone is worried about Chinese models, but the frontier labs was expected to be fine; we need to enable open U.S.
The new ggml-based library is designed to embed local transcription models across macOS, Windows and Linux while validating output against reference implementations.
After selling 2,500 MIDI recorders, developer Chip Weinberger says the device’s software demanded more effort than its deliberately simple physical design.
Testing found that connecting certain LG displays could trigger an app installer that repeatedly promoted a McAfee trial without a separate approval prompt.
A personal trial found similar practical coding output from Kimi and Claude, but its conclusions rely on one user’s workload rather than controlled benchmarking.
The maiden flight carried a varied set of small payloads and marked a major test of India's effort to expand private participation in its space industry.
Current and former nurses say surveillance metrics can discourage longer patient conversations; Kaiser says its quality tools include human oversight and do not use average handle time for assessment.
The downloadable typeface overlays high- and low-frequency letterforms so people and vision models may perceive different messages, though its creator warns it is not secure protection.
The multimodal mixture-of-experts model offers a one-million-token context window and is available through Kimi services, with weights promised by July 27.
Observations of the nearby super-Earth provide the first direct atmospheric detection for a rocky world in a star’s habitable zone, while leaving its water and prospects for life unresolved.