Opus
Coverage of Opus in the Nexus archive.
- Andrej Karpathy’s trick for getting better answers from AI: Just ramble at it for 10 minutes
Andrej Karpathy, an AI researcher at Anthropic, recommends rambling for 10 minutes in voice mode to help large language models (LLMs) extract key information. This approach allows humans to unload unstructured thoughts, which the AI then organizes. Companies like OpenAI and Anthropic are advancing voice agent models and hardware to facilitate such interactions.
- Claude’s voice mode is now available for Opus and Sonnet
Anthropic has expanded voice mode availability to its Opus and Sonnet models, previously only on Haiku. The update extends voice mode to apps like Gmail, Slack, and Canva, with users increasingly applying it to business problem-solving rather than casual queries.
- GLM 5.2 vs. Opus
The article compares GLM 5.2 and Opus, two models, with a Hacker News discussion thread showing 17 points and 4 comments.
- Elevated error rates on Opus 4.7
Elevated error rates have been reported on Opus 4.7, with an article and comments available on the status page and news website. The incident has 1 comment and 4 points. The issue is related to technology.
- We decreased our LLM costs with Opus
Mendral reduced their large language model (LLM) costs by implementing Opus, a cost-optimization tool for AI infrastructure. The article highlights their technical approach to lowering expenses without compromising performance.
- Opus 4.7 to 4.6 Inflation is ~45%
The article reports a slight decrease in Opus inflation from 4.7% to 4.6%, though the rate remains around 45%. It includes a link to the article and Hacker News comments with 88 points and 64 discussions.