OpenAI has released GPT-5.2 Codex to developers through the Responses API. The model was previously limited to the Codex environment. According to OpenAI Developers, it excels at complex, tedious tasks like developing new features, refactoring code, and tracking down bugs. OpenAI also says it's their best cybersecurity model yet, helping identify vulnerabilities in codebases.
The model accepts text and images as input and offers four levels of reasoning effort: low, medium, high, and very high. Pricing comes in at $1.75 per million input tokens and $14 per million output tokens, a notable increase from earlier GPT-5 Codex models, which cost $1.25 per million input tokens and $10 per million output tokens.
Coding platforms Cursor and Windsurf have already integrated the model, with Windsurf offering it at half price for a limited time. OpenAI has published a prompting guide.
Microsoft has also told its Azure sales team to count Anthropic model sales to cloud customers toward their quotas, just like they would for Microsoft's own software. That's unusual for third-party products, which typically generate less revenue for Azure. The deeper collaboration follows Microsoft's investment of up to $5 billion in Anthropic last November.
Google taps its massive data advantage with new Gemini feature
Google knows where you went on vacation, what you bought, and who you email. Now that knowledge is supposed to make your AI assistant smarter. The new “Personal Intelligence” feature connects Gemini with Gmail, Google Photos, and YouTube—an advantage competitors can’t match, if it works as intended.
Deutsche Telekom is soon using AI voice agents from Elevenlabs in its customer service. Customers can talk to realistic-sounding AI voices around the clock through the app or by phone. The partnership between Europe's largest telecom company and the AI audio startup goes back a while. Since October 2025, Magenta customers have been able to convert text into podcasts up to 25 times a month for free in the MeinMagenta app. Deutsche Telekom also invested in Elevenlabs' Series C funding round.
According to Elevenlabs' internal data, the AI support agent successfully resolves about 80 percent of user queries, particularly when it comes to specific documentation questions. For more complex issues like troubleshooting or pricing inquiries, though, the system still hits its limits and needs to hand off to human support.
Apple's expanded partnership with Google relegates ChatGPT to a backup option on iPhone.
An Apple spokesperson told The Information that deeper Gemini integration would not affect the company's agreement with OpenAI to use ChatGPT for more complex requests.
For OpenAI, though, the Gemini-Siri integration marks the end of a failed partnership. CEO Sam Altman and his team originally hoped the Apple deal would bring in billions in additional revenue and break Google's dominance on Apple devices.
That didn't happen. According to a source speaking to The Information, the partnership brought OpenAI almost no additional traffic. Now Google's model handles everyday user interactions, while ChatGPT serves only as an optional fallback.
Google Deepmind has updated Veo 3.1 with new features for generating video from reference images. The update enables more dynamic and expressive videos, even with simple text prompts, according to Google. Users can now keep characters consistent across multiple scenes and seamlessly combine different elements like textures, objects, and backgrounds.
The updates are available now in the Gemini app, YouTube, Flow, Google Vids, and through the Gemini API and Vertex AI. All generated videos include the invisible SynthID watermark to identify AI-generated content. Users can check in the Gemini app whether a video was created with Google AI.
China reportedly tightens Nvidia H200 restrictions, limits purchases to special cases
The AI race between the US and China enters a new phase: Washington loosens Nvidia export rules, but Beijing reportedly halts purchases. China wants to shield its chip industry and may require buyers to also purchase domestic chips.
Salesforce has launched a new Slackbot built on Anthropic's Claude AI model. According to co-founder and Chief Technology Officer Parker Harris, the company is also testing alternatives. The AI assistant lives directly inside Slack and can search data across Slack, Salesforce, Google Drive, Box, and Atlassian's Confluence. It uses context from conversations, files, and channels to answer questions, create content, and prepare meetings, while respecting existing access rights and permissions.
Salesforce
Slackbot is now available for Business+ and Enterprise+ customers, with a gradual rollout running through February. Down the road, Slackbot will also be able to work with Agentforce and other AI agents.
A new leak reveals details about OpenAI's planned hardware, an audio device designed to compete with Apple's Airpods.X and Weibo leaker "Smart Pikachu" claims OpenAI is developing a device codenamed "Sweetpea" with designer Jony Ive reportedly involved. The alleged September launch targets 40 to 50 million units in year one.
The device supposedly features an oval metal housing with two capsule-shaped components worn behind the ear, running on a 2nm chip with Samsung Exynos as the frontrunner. A separate chip would enable iPhone control through Siri. Material costs are reportedly close to smartphone level.
The leaked diagram shows "Sweetpea's" alleged components: EMG signal window, ceramic skin contact window, mainboard with lithium-ion battery, and ultrasonic transmitter/receiver module. | Image: via zhihuipikachu
If the leak proves accurate, Foxconn could produce up to five OpenAI devices by 2028, including a pen codenamed "Gumdrop." The manufacturer reportedly sees this as a chance to recover after losing all Airpods programs to Luxshare. OpenAI allegedly favored Luxshare initially but switched to Foxconn to enable production outside China.
