AI News
AI community posts, news and release updates
Latest updates
- News
ChatGPT Voice Now Available in Codex via iOS with Remote Access, Android Coming Soon
Published Jul 23, 2026OpenAI announced that ChatGPT Voice can now be used in Codex from the iOS app with paired remote access. Android support is on the way.
- ChatGPT
- codex
- Android
- iOS
- voice
Some are exploiting a trick: encoding prompts as PNG images to slash token usage in Claude Code, potentially saving up to 70%.
- Claude Code
- 省钱
- PNG提示
A new tool lets you run Claude Code inside multiple AI terminals like Codex, Kimi, Grok, and Cursor, expanding its reach.
- Cursor
- Claude Code
- codex
- Kimi
- Grok
Grok 4.5 scores 85.7% on ARC-AGI-1 at $0.33/task, 52.6% on ARC-AGI-2 at $0.78/task, and 0.3% on ARC-AGI-3 costing $6.9K. Increasing reasoning effort didn't improve performance.
- Grok
- 基准测试
- xAI
- ARC-AGI
- Rant
LOL! US government as LLM benchmark authority? Kimi K3 called far worse than frontier models
Published Jul 23, 2026The US government now claims Kimi K3 is much worse than US frontier models, and LiveBench AI agrees. The author snarks that Kimi is just a cheap Sonnet/Opus 4.6 class model for long-running tasks, not frontier intelligence.
- LLM benchmark
- US government
- Kimi K3
- News
SF digital billboards show AI intelligence vs. cost: Fable 5 leads but pricey
Published Jul 23, 2026Artificial Analysis posted a chart on SF digital billboards comparing AI intelligence index vs. cost per task, plotting the Pareto frontier. Fable 5 scores 60 but costs $2.75/task; Grok 4.5 (high) scores 54 at $0.31, ~9x cheaper; DeepSeek v4 Flash scores 44 at $0.04, ~69x cheaper than Fable.
- Grok
- DeepSeek
- AI模型
- Fable 5
- 成本效率
- Pareto前沿
Open-artifact lets you easily view, manage, and share documents and artifacts generated by various LLMs, enabling collaboration.
- LLM
- Open-artifact
- 文档分享
DingDuff is an MCP connector that enables Claude to access legal databases and research tools, assisting legal workflows.
- MCP
- Claude
- DingDuff
- 法律研究
DeepSeek founder Liang Wenfeng broke his long silence at an investor meeting, sharing the company's strategy and views on AI development.
- DeepSeek
- 梁文锋
- AI行业动态
- Launch
Start new tasks, check existing threads, and steer work across Codex and ChatGPT Work
Published Jul 23, 2026Describes a new capability in Codex and ChatGPT Work to start tasks, check threads, and steer work across platforms in one conversation.
- codex
- tasks
- ChatGPT Work
- threads