Grok 4.5 makes the AI race about efficiency

AI companies are no longer just racing to build smarter models; they are racing to make them cheaper, safer, and easier to use. SpaceXAI is positioning Grok 4.5 around speed and cost. Anthropic is testing a way to switch off risky knowledge inside models. OpenAI is pushing voice closer to a natural computing interface. Together, these updates show where AI is heading next: not just more intelligence, but more control over how that intelligence is used.

In today’s post:

• Grok 4.5 is competing on cost

• Voice may be AI’s next big interface

• Anthropic is testing a sharper way to control risky AI knowledge.

SPONSORED BY

HR and IT need to work as one. Here's how

Onboarding, offboarding, role changes, leave—every employee lifecycle moment requires HR and IT to move together. When they don't, people fall through the cracks. Access delays mount. Compliance risk creeps in.

This guide gives HR and IT leaders a practical communication framework to close the gaps, standardize handoffs, and keep the employee experience seamless from day one to last day. Free download—built for ops teams that need it to actually work.

What’s Trending Today

LAUNCH

SpaceXAI is betting that cheaper intelligence can win attention

SpaceXAI just released Grok 4.5. Elon Musk is calling it an “Opus-class” model. That is a bold comparison. But the more interesting claim is not power. It is efficiency.

Here's everything you need to know:

• SpaceXAI says Grok 4.5 can handle coding, writing, research, and office work.

• The company claims it uses tokens twice as efficiently as leading models.

• That matters because token cost is becoming a real business concern.

• Grok 4.5 is priced at $2 per million input tokens.

• Its output tokens cost $6 per million, which is aggressive.

• Musk says it is comparable to Anthropic’s Opus 4.7, but faster.

• The real test will be whether users feel that difference in daily work.

AI model launches are starting to look different. The old question was simple. Which model is smartest? Now the question is broader. Which model is smart enough, fast enough, and cheap enough? That shift matters. Most companies do not need perfection. They need reliable output at a cost they can scale. If Grok 4.5 delivers there, it becomes harder to ignore.

BREAKTHROUGH

OpenAI is making ChatGPT feel less like typing and more like talking

OpenAI just released GPT-Live-1 and GPT-Live-1 mini. These models are built for live conversation. They can listen and speak at the same time. That makes interruptions feel more natural. It also makes AI feel less like a tool. And more like something you work with in real time.

Here's everything you need to know:

• OpenAI says the new voice models handle turn-taking better than before.

• GPT-Live-1 mini will replace Advanced Voice Mode by default in ChatGPT.

• Paid users will get access to the larger GPT-Live-1 model.

• The models can pass tougher queries to newer GPT systems while talking.

• They can stay silent, absorb context, and respond when needed.

• OpenAI sees voice becoming a primary interface for complex work.

• The demo still showed limits, especially with natural Hindi translation.

It changes the relationship people have with software. Typing makes AI feel like a search box. Voice makes it feel like a collaborator. But that also raises the bar. People expect conversation to be smooth, patient, and natural. One awkward pause or strange accent breaks the illusion. The winners here will not just build smarter models. They will build assistants that know when not to speak.

STRATEGY

AI models may need an off switch

Anthropic just shared early research on something called GRAM. The idea is simple, but important. Some AI knowledge can help or harm. Cybersecurity can protect systems, or attack them. Virology can support vaccines, or enable dangerous misuse. So the question becomes: can models “forget” risky knowledge when needed?

Here's everything you need to know:

• Current safety systems mostly block dangerous outputs, not the knowledge inside the model.

• A determined attacker may still try to bypass refusals and filters.

• GRAM tries a different approach by storing risky knowledge in removable modules.

• Each module handles one category, like virology, cybersecurity, or nuclear physics.

• When a module is removed, the related capability becomes harder to access.

• Anthropic found this worked similarly to training separate filtered models.

• The early tests also showed little damage to general model performance.

This feels like a more realistic direction for AI safety. Not because it solves everything. It does not. Some knowledge may be too connected to remove cleanly. But it shifts safety from surface-level blocking to deeper access control. That matters. The future may not be one model for everyone. It may be one model with different knowledge switches. And the hard part will be deciding who gets which switch.

Free Guides

My Free Guides to Download:

🚀 Founders & AI Builders, Listen up!

If you’ve built an AI tool, here’s an opportunity to gain serious visibility.

Nextool AI is a leading tools aggregator that offers:

• 500k+ page views and a rapidly growing audience.

• Exposure to developers, entrepreneurs, and tech enthusiasts actively searching for innovative tools.

• A spot in a curated list of cutting-edge AI tools, trusted by the community.

• Increased traffic, users, and brand recognition for your tool.

Take the next step to grow your tool’s reach and impact.

Reach 150,000+ READERS:

Expand your reach and boost your brand’s visibility!

Partner with Nextool AI to showcase your product or service to 140,000+ engaged subscribers, including entrepreneurs, tech enthusiasts, developers, and industry leaders.

Ready to make an impact? Visit our sponsorship website to explore sponsorship opportunities and learn more!