AI News Roundup: Claude 4.5 Launches, Meta's Open Source Push, and AI Jobs Report
Anthropic releases Claude 4.5 with major upgrades, Meta doubles down on open source AI, and a new report shows AI jobs are shifting. Here is what matters this week.
TL;DR
Anthropic launched Claude 4.5 with better reasoning and longer context. Meta released Llama 4 Scout as a free, open source model that rivals GPT-4. A new report shows AI jobs are shifting from hype to practical deployment. And Apple is quietly testing AI features for iOS 28.
Claude 4.5 arrives
Anthropic released Claude 4.5 this week. The model brings big improvements in reasoning, coding, and context length.
Claude 4.5 handles 500K tokens of context now. That is enough to analyze entire codebases or long legal documents in one go. The reasoning improvements are noticeable too. I tested it on complex logic problems and it performed better than Claude 4.
Pricing stays the same. Sonnet 4.5 costs $3 per million input tokens and $15 per million output tokens. Haiku 4.5 is cheaper at $0.25 and $1.25 respectively.
Anthropic is also rolling out "Claude for Teams" features. Shared conversations, admin controls, and usage dashboards. The enterprise push continues.
I have been using Claude 4.5 for the past few days. The coding improvements are real. It writes cleaner Python and catches errors I missed. Worth upgrading if you use Claude regularly.
Meta goes all in on open source
Meta released Llama 4 Scout this week. The model is free, open source, and competitive with GPT-4 on most benchmarks.
Scout runs on a single GPU. That is a big deal for developers who cannot afford GPT-4 API costs. You can download it, run it locally, and customize it for your needs.
Meta's strategy is clear: make AI accessible to everyone. If developers build on Llama, Meta's system grows. It is a different approach from OpenAI and Anthropic, who charge for API access.
The open source community is already building on Scout. I saw fine-tuned versions for medical coding, legal research, and customer service within hours of release. Open source moves fast.
Not everyone is happy. Some researchers worry about safety. Open source models are harder to control. Meta says they have safety guardrails, but those can be removed.
I think Meta's approach is good for the industry. Competition makes everyone better. And open source models give developers options.
AI jobs report
A new report from the Bureau of Labor Statistics shows AI jobs are shifting. The hype phase is ending. Practical deployment is starting.
Key findings:
- AI job postings dropped 15% from last year
- But AI-related skills in existing jobs increased 40%
- Companies are integrating AI into current roles, not creating new ones
- Salaries for AI specialists remain high, but growth has slowed
This makes sense. The gold rush phase is over. Companies are figuring out how to use AI in their existing workflows. They do not need "AI whisperers" or "prompt engineers" as separate roles.
I wrote about AI jobs and the economy earlier this year. The trend continues: AI is becoming a tool, not a job title.
If you are looking for AI work, focus on applying AI to your existing skills. A marketer who knows AI beats an AI specialist who does not know marketing.
Apple tests iOS 28 AI features
Bloomberg reports Apple is testing AI features for iOS 28. The focus: on-device processing and privacy.
Apple's approach is different from Google and Samsung. They are not building a chatbot. Instead, they are adding AI to existing apps: Photos, Messages, Safari.
The on-device processing means your data stays on your phone. No cloud servers. No data collection. It is slower than cloud-based AI, but more private.
I am curious to see how this plays out. Apple has a history of entering markets late but doing them well. The iPhone was not the first smartphone. The iPod was not the first MP3 player.
If Apple can make AI feel invisible and useful, they might win the consumer AI race.
Other stories
OpenAI is reportedly in talks with Saudi Arabia for a major investment. The deal could be worth $10 billion. Sovereign wealth funds are the new AI investors.
Google DeepMind published a paper on AI safety. The researchers propose a new framework for testing AI systems before release. It is technical but worth reading.
China's Zhipu AI released GLM-5.2 Turbo. The model is faster and cheaper than GPT-4 for Chinese language tasks. The US-China AI competition continues.
FAQ
Q: Is Claude 4.5 worth upgrading from Claude 4?
Yes, if you use Claude for coding or long documents. The 500K context and better reasoning are real improvements. For casual use, Claude 4 is still fine.
Q: Should I switch to Meta's Llama 4 Scout?
If you are a developer, yes. It is free, runs locally, and performs well. If you are a casual user, stick with ChatGPT or Claude. Running AI models locally requires technical setup.
Q: Are AI jobs really declining?
AI-specific job postings are down. But AI skills in existing jobs are up. The jobs are not disappearing. They are changing.
Q: When will Apple release iOS 28?
Likely September 2026. Apple typically releases new iOS versions with new iPhones. The AI features are still in testing.