AI News
AI community posts, news and release updates
Latest updates
A context engineering platform that self-improves has debuted, aiming to enhance how AI models utilize context and reduce manual tuning.
- context engineering
- AI platform
- self-improving
- Discussion
Opus 5 Might as Well Be Fable 6: User Tests Show It Completely Destroys Fable 5
Published Jul 24, 2026A user enthuses that Claude Opus 5 is essentially Fable 6, as it utterly outclasses Fable 5 in their testing. The difference is described as 'ridiculous'.
- comparison
- Performance
- Claude Opus 5
This article summarizes context engineering rules for Claude 5 models, helping developers optimize prompts and context structure for better output.
- Prompt Engineering
- context engineering
- Claude 5
- Discussion
When is an apology not an apology? From an AI boss with a rogue chatbot
Published Jul 24, 2026An article about an AI company boss apologizing for a chatbot's misbehavior, questioning the line between genuine apology and PR spin.
- ethics
- AI boss
- apology
- rogue chatbot
Grok 4.5 reduced its cache token price from $0.50 to $0.30 per 1M tokens, cutting costs for developers.
- API
- Grok
- 降价
People are only beginning to explore how ChatGPT can be used. We believe AI with agentic abilities will help people live better.
- ChatGPT
- Agentic AI
- empowerment
- News
Claude Opus 5 tops Intelligence Index, matches Fable 5 at 26% lower cost per task
Published Jul 24, 2026Artificial Analysis evaluates Claude Opus 5: scores 61 on Intelligence Index, tied with Fable 5, with 26% lower cost per task. Leads in agentic knowledge work and coding, but hallucination rate rises to 50%.
- cost
- benchmark
- agentic
- Artificial Analysis
- Claude Opus 5
Ramp tested AI models on 150k real business invoices. Grok 4.5 achieved the highest perfect-extraction rate, outperforming Gemini Flash 3.6, GPT 5.6 Terra, and Sonnet 5. The task requires long-context reasoning over 100K+ tokens.
- 长上下文
- AI比较
- Grok 4.5
- Ramp
- 发票处理
- News
Jensen Huang: US should not restrict open-weight AI models; 25 companies sign letter
Published Jul 24, 2026Nvidia CEO Jensen Huang made his first-ever X post to share a letter signed by 25 companies (Meta, Microsoft, IBM, Palantir, etc.) urging the US not to restrict open-weight AI models. OpenAI and Anthropic were absent. The letter follows renewed debate sparked by China's Kimi K3 model.
- Nvidia
- policy
- Jensen Huang
- Kimi K3
- open-weight AI
Someone questioned if Kimi K3 surpasses Opus, but the release of Opus 5 reversed the situation, sparking discussions on model performance.
- model comparison
- Opus
- Kimi K3