Daily AI News | OpenAI releases GPT-6 Sol and Luna with lower API prices
# Daily AI News | OpenAI releases GPT-6 Sol and Luna with lower API prices
## Lead Story
### OpenAI releases GPT-6 Sol and Luna with lower API prices
OpenAI released GPT-6 Sol and Luna as faster, lower-cost choices alongside GPT-6 Astra. Its listed prices per million input and output tokens are $2 and $10 for Sol, and $0.10 and $0.50 for Luna. Both are available through the API, ChatGPT Work and Codex; they are not yet in the regular Chat interface, and the product rollout is gradual.
**AI take:** Developers gain cheaper model options, while access in each product still depends on the rollout.
## More News
### Anthropic releases Claude Opus 5.5
Anthropic released Claude Opus 5.5, the first model in its Claude 5.5 family, across its platform and AWS, Google Cloud and Azure. List prices are $4 per million input tokens and $20 per million output tokens, 20% below Opus 5. Anthropic says lower token use can reduce costs for typical workloads by about 40%, though results depend on the task.
**AI take:** Complex workloads may become cheaper, but buyers should calculate costs against their own usage.
### NVIDIA releases Isaac ROS 5.0 for robotics developers
NVIDIA released the free, open-source Isaac ROS 5.0 at ROSCon. It adds skills and documentation prepared for AI-assisted robotics development, supports ROS Lyrical and Ubuntu 24.04, and provides workflows for setup, perception and manipulation. The release is available now for teams building and deploying robots on Jetson hardware.
**AI take:** Ready-made components may shorten integration work, although each robot still needs real-world testing.
### Microsoft and partners disrupt AI-assisted phishing service EvilTokens
Microsoft said it and its partners seized 50 websites used by EvilTokens and disabled more than 150 related domains. The service used AI to analyze compromised inboxes, identify payment relationships and help prepare fraud messages. Microsoft links it to more than 12,000 compromised inboxes at over 10,000 organizations worldwide. British police earlier arrested two men on suspicion of related offenses; the investigation continues.
**AI take:** The takedown does not erase similar risks, so organizations still need to verify payment changes and revoke stolen sessions.
### Qualcomm unveils two flagship phone chips focused on on-device AI
Qualcomm introduced the sixth-generation Snapdragon 8 Elite and a higher-end Extreme variant at its Snapdragon Summit, emphasizing AI tasks that run on phones. The company says the top chip can run a roughly 30-billion-parameter mixture-of-experts model locally. Phone makers still need to turn that capability into shipping devices and software features.
**AI take:** The on-device compute ceiling rises, but battery life, speed and useful features need testing in actual phones.
### Runway launches DIFFUSE for AI-native creative talent
Runway launched DIFFUSE so brands and agencies can find and contact AI-native creators and studios. The live site shows talent profiles and project examples. Its jobs feature is labeled as coming soon, so a complete hiring workflow is not yet established.
**AI take:** Creators gain another place to be discovered, while sustained hiring activity remains to be demonstrated.
### OpenAI adds prompt-cache diagnostics for GPT-6 developers
OpenAI separately released GPT-6 prompt-caching improvements, saying default hit rates are higher and adding a dashboard, cache-miss diagnostics and controls for choosing reusable prompt prefixes. Eligible repeated input can receive a cached-read discount of up to 90%, according to OpenAI. Actual savings depend on context reuse and the billing mix.
**AI take:** Long-running apps can now identify cache misses, but the discount is not a direct reduction in the whole bill.
## Rumors
