Claude Opus 5: Frontier AI at half price

Twitter

Claude Opus 5's counterintuitive FrontierCode performance decline at higher reasoning effort is caused by the benchmark penalizing out-of-scope improvements, revealing how evaluation design can misalign with actual model capabilities. Despite this specific quirk, Opus 5 demonstrates superior coding performance at higher effort levels across most other benchmarks.

VentureBeat

OpenAI brings full-duplex GPT-Live voice control to its desktop ChatGPT app, enabling developers to manage complex coding tasks hands-free through Codex and ChatGPT Work. The integration allows simultaneous multi-task execution via natural voice commands, though it remains limited to paid enterprise subscribers under a proprietary license.

Google Blog

Google's Q2 2026 earnings showed exceptional AI-driven growth with Cloud revenue up 82% and $514 billion in backlog, powered by Gemini Enterprise adoption across nearly 90% of Fortune 100 companies and surging demand for AI infrastructure that has left the company supply-constrained while processing 22 billion tokens per minute.

Anthropic

Anthropic launched Claude Opus 5, delivering Fable-level AI performance at half the cost with state-of-the-art results on coding benchmarks and three times better performance than competitors on novel problem-solving tasks. The model also represents Anthropic's most aligned release, with the lowest rates of reckless behavior and strong adherence to safety principles.

Meta AI

Meta's open-source SAM 3 and DINOv3 models are powering the DOE's SYNAPS-I project, reducing scientific image analysis from one month to 15 minutes and enabling real-time discovery at national labs handling petabytes of experimental data. The open-source approach allows secure deployment on government infrastructure while fine-tuning for specialized scientific applications.